| 1 | //! Voice input commands — `/voice`, `/voice-send`, `/voice-control`. |
| 2 | //! |
| 3 | //! Records audio from the default microphone, sends it to the configured |
| 4 | //! provider's API for transcription, and inserts the transcribed text into |
| 5 | //! the composer. The interaction model mirrors MiMo Code's voice UX: |
| 6 | //! |
| 7 | //! `/voice` — toggle voice input on/off (records when toggled on) |
| 8 | //! `/voice-send` — toggle auto-send when the transcript ends with |
| 9 | //! "send it" / "发送" |
| 10 | //! `/voice-control` — toggle AI-assisted dictation that sees the current |
| 11 | //! composer text |
| 12 | //! |
| 13 | //! The slash commands only flip state and emit [`AppAction::VoiceCapture`]; |
| 14 | //! the actual capture runs in the UI event loop where the live [`Config`] |
| 15 | //! supplies provider credentials. That keeps the handlers side-effect free |
| 16 | //! (the registry smoke tests execute every command) and avoids caching |
| 17 | //! auth material on [`App`]. |
| 18 | //! |
| 19 | //! ## Recording |
| 20 | //! |
| 21 | //! Uses platform-specific command-line tools (sox, rec, arecord) to capture |
| 22 | //! 16kHz mono 16-bit PCM audio. Records until a silence gap is detected or |
| 23 | //! the maximum duration is reached (default 10 s). |
| 24 | |
| 25 | use std::process::{Command, Stdio}; |
| 26 | use std::sync::LazyLock; |
| 27 | use std::time::Duration; |
| 28 | |
| 29 | use regex::Regex; |
| 30 | |
| 31 | use crate::commands::CommandResult; |
| 32 | use crate::commands::traits::{CommandInfo, RegisterCommand}; |
| 33 | use crate::config::Config; |
| 34 | use crate::tui::app::{App, AppAction}; |
| 35 | use codewhale_localization::{MessageId, tr}; |
| 36 | |
| 37 | /// Transcription model requested from the provider's chat-completions API. |
| 38 | const ASR_MODEL: &str = "mimo-v2.5-asr"; |
| 39 | const GROQ_ASR_MODEL: &str = "whisper-large-v3-turbo"; |
| 40 | /// Local whisper binary names to probe (whisper.cpp, faster-whisper, OpenAI whisper). |
| 41 | const LOCAL_WHISPER_BINS: &[&str] = &["whisper", "whisper.cpp", "whisper-cpp", "faster-whisper"]; |
| 42 | /// Model used for the AI-assisted voice-control pipeline. |
| 43 | const VOICE_CONTROL_MODEL: &str = "mimo-v2.5"; |
| 44 | |
| 45 | pub(in crate::commands) const VOICE_INFO: CommandInfo = CommandInfo { |
| 46 | name: "voice", |
| 47 | aliases: &["yuyin", "语音"], |
| 48 | usage: "/voice", |
| 49 | description_id: MessageId::CmdVoiceDescription, |
| 50 | }; |
| 51 | |
| 52 | pub(in crate::commands) const VOICE_SEND_INFO: CommandInfo = CommandInfo { |
| 53 | name: "voicesend", |
| 54 | aliases: &["voice-send", "yuyinsend", "语音发送"], |
| 55 | usage: "/voicesend", |
| 56 | description_id: MessageId::CmdVoiceSendDescription, |
| 57 | }; |
| 58 | |
| 59 | pub(in crate::commands) const VOICE_CONTROL_INFO: CommandInfo = CommandInfo { |
| 60 | name: "voicecontrol", |
| 61 | aliases: &["voice-control", "yuyincontrol", "语音控制"], |
| 62 | usage: "/voicecontrol", |
| 63 | description_id: MessageId::CmdVoiceControlDescription, |
| 64 | }; |
| 65 | |
| 66 | pub(in crate::commands) struct VoiceCmd; |
| 67 | pub(in crate::commands) struct VoiceSendCmd; |
| 68 | pub(in crate::commands) struct VoiceControlCmd; |
| 69 | |
| 70 | impl RegisterCommand for VoiceCmd { |
| 71 | fn info() -> &'static CommandInfo { |
| 72 | &VOICE_INFO |
| 73 | } |
| 74 | |
| 75 | fn execute(app: &mut App, _arg: Option<&str>) -> CommandResult { |
| 76 | voice(app) |
| 77 | } |
| 78 | } |
| 79 | |
| 80 | impl RegisterCommand for VoiceSendCmd { |
| 81 | fn info() -> &'static CommandInfo { |
| 82 | &VOICE_SEND_INFO |
| 83 | } |
| 84 | |
| 85 | fn execute(app: &mut App, _arg: Option<&str>) -> CommandResult { |
| 86 | voice_send(app) |
| 87 | } |
| 88 | } |
| 89 | |
| 90 | impl RegisterCommand for VoiceControlCmd { |
| 91 | fn info() -> &'static CommandInfo { |
| 92 | &VOICE_CONTROL_INFO |
| 93 | } |
| 94 | |
| 95 | fn execute(app: &mut App, _arg: Option<&str>) -> CommandResult { |
| 96 | voice_control(app) |
| 97 | } |
| 98 | } |
| 99 | |
| 100 | // --- Recorder detection ---------------------------------------------------- |
| 101 | |
| 102 | /// Platform-specific recorder definitions. |
| 103 | #[derive(Debug, Clone)] |
| 104 | struct Recorder { |
| 105 | cmd: &'static str, |
| 106 | /// CLI arguments for piping raw 16kHz mono S16_LE PCM to stdout. |
| 107 | pipe_args: &'static [&'static str], |
| 108 | } |
| 109 | |
| 110 | fn detect_recorder() -> Option<Recorder> { |
| 111 | // Operator kill-switch: a headless `serve --http` host has no business |
| 112 | // opening a microphone; disabling voice here makes `GET /v1/voice` |
| 113 | // report `available: false` and every dictate call fail closed. |
| 114 | if std::env::var_os("CODEWHALE_DISABLE_VOICE").is_some() { |
| 115 | return None; |
| 116 | } |
| 117 | let candidates: &[Recorder] = if cfg!(target_os = "macos") { |
| 118 | &[ |
| 119 | Recorder { |
| 120 | cmd: "sox", |
| 121 | pipe_args: &["-d", "-r", "16000", "-c", "1", "-b", "16", "-t", "raw", "-"], |
| 122 | }, |
| 123 | Recorder { |
| 124 | cmd: "rec", |
| 125 | pipe_args: &["-r", "16000", "-c", "1", "-b", "16", "-t", "raw", "-"], |
| 126 | }, |
| 127 | ] |
| 128 | } else if cfg!(target_os = "linux") { |
| 129 | &[ |
| 130 | Recorder { |
| 131 | cmd: "arecord", |
| 132 | pipe_args: &["-f", "S16_LE", "-r", "16000", "-c", "1", "-t", "raw"], |
| 133 | }, |
| 134 | Recorder { |
| 135 | cmd: "sox", |
| 136 | pipe_args: &["-d", "-r", "16000", "-c", "1", "-b", "16", "-t", "raw", "-"], |
| 137 | }, |
| 138 | ] |
| 139 | } else if cfg!(target_os = "windows") { |
| 140 | &[Recorder { |
| 141 | cmd: "sox", |
| 142 | pipe_args: &["-d", "-r", "16000", "-c", "1", "-b", "16", "-t", "raw", "-"], |
| 143 | }] |
| 144 | } else { |
| 145 | &[] |
| 146 | }; |
| 147 | |
| 148 | candidates |
| 149 | .iter() |
| 150 | .find(|r| { |
| 151 | Command::new(r.cmd) |
| 152 | .arg("--version") |
| 153 | .stdin(Stdio::null()) |
| 154 | .stdout(Stdio::null()) |
| 155 | .stderr(Stdio::null()) |
| 156 | .spawn() |
| 157 | .is_ok() |
| 158 | }) |
| 159 | .cloned() |
| 160 | } |
| 161 | |
| 162 | /// Check whether voice recording is available on this system. |
| 163 | pub fn is_available() -> bool { |
| 164 | detect_recorder().is_some() |
| 165 | } |
| 166 | |
| 167 | // --- WAV encoding ---------------------------------------------------------- |
| 168 | |
| 169 | /// Encode raw 16kHz mono S16_LE PCM samples as a WAV buffer. |
| 170 | fn encode_wav(samples: &[i16]) -> Vec<u8> { |
| 171 | let data_size = (samples.len() * 2) as u32; |
| 172 | let sample_rate: u32 = 16000; |
| 173 | let mut buf = Vec::with_capacity(44 + data_size as usize); |
| 174 | |
| 175 | // RIFF header |
| 176 | buf.extend_from_slice(b"RIFF"); |
| 177 | buf.extend_from_slice(&(36 + data_size).to_le_bytes()); |
| 178 | buf.extend_from_slice(b"WAVE"); |
| 179 | |
| 180 | // fmt chunk |
| 181 | buf.extend_from_slice(b"fmt "); |
| 182 | buf.extend_from_slice(&16u32.to_le_bytes()); // chunk size |
| 183 | buf.extend_from_slice(&1u16.to_le_bytes()); // PCM |
| 184 | buf.extend_from_slice(&1u16.to_le_bytes()); // mono |
| 185 | buf.extend_from_slice(&sample_rate.to_le_bytes()); |
| 186 | buf.extend_from_slice(&(sample_rate * 2).to_le_bytes()); // byte rate |
| 187 | buf.extend_from_slice(&2u16.to_le_bytes()); // block align |
| 188 | buf.extend_from_slice(&16u16.to_le_bytes()); // bits per sample |
| 189 | |
| 190 | // data chunk |
| 191 | buf.extend_from_slice(b"data"); |
| 192 | buf.extend_from_slice(&data_size.to_le_bytes()); |
| 193 | for &sample in samples { |
| 194 | buf.extend_from_slice(&sample.to_le_bytes()); |
| 195 | } |
| 196 | |
| 197 | buf |
| 198 | } |
| 199 | |
| 200 | // --- Recording ------------------------------------------------------------- |
| 201 | |
| 202 | /// Maximum recording duration in seconds before auto-stopping. |
| 203 | pub const MAX_RECORD_SECS: u64 = 10; |
| 204 | /// Minimum segment duration in seconds to consider as valid speech. |
| 205 | const MIN_SEGMENT_SECS: f64 = 0.3; |
| 206 | |
| 207 | /// Record audio from the default microphone. |
| 208 | /// |
| 209 | /// Returns raw 16kHz mono S16_LE PCM samples. Returns `None` if no recorder |
| 210 | /// is available, the recording failed, or no speech was detected. |
| 211 | fn record_audio() -> Option<(Vec<i16>, Duration)> { |
| 212 | let recorder = detect_recorder()?; |
| 213 | let start = std::time::Instant::now(); |
| 214 | |
| 215 | let mut child = Command::new(recorder.cmd) |
| 216 | .args(recorder.pipe_args) |
| 217 | .stdin(Stdio::null()) |
| 218 | .stdout(Stdio::piped()) |
| 219 | .stderr(Stdio::null()) |
| 220 | .spawn() |
| 221 | .ok()?; |
| 222 | |
| 223 | let stdout = child.stdout.take()?; |
| 224 | let mut reader = std::io::BufReader::new(stdout); |
| 225 | let mut all_samples: Vec<i16> = Vec::with_capacity(16000 * MAX_RECORD_SECS as usize); |
| 226 | |
| 227 | // Read until timeout or silence |
| 228 | let mut buf = [0u8; 320]; // 10ms of 16kHz S16_LE |
| 229 | let max_duration = Duration::from_secs(MAX_RECORD_SECS); |
| 230 | let mut silence_samples = 0u32; |
| 231 | let mut had_speech = false; |
| 232 | let speech_threshold: i16 = 500; // RMS-based speech detection threshold |
| 233 | let silence_duration_samples = 16000u32; // 1 second of silence to stop |
| 234 | |
| 235 | loop { |
| 236 | use std::io::Read; |
| 237 | match reader.read_exact(&mut buf) { |
| 238 | Ok(()) => { |
| 239 | let chunk: Vec<i16> = buf |
| 240 | .as_chunks::<2>() |
| 241 | .0 |
| 242 | .iter() |
| 243 | .copied() |
| 244 | .map(i16::from_le_bytes) |
| 245 | .collect(); |
| 246 | |
| 247 | // Simple RMS-based VAD |
| 248 | let rms = (chunk.iter().map(|&s| (s as f64) * (s as f64)).sum::<f64>() |
| 249 | / chunk.len() as f64) |
| 250 | .sqrt(); |
| 251 | let is_speech = rms > speech_threshold as f64; |
| 252 | |
| 253 | if is_speech { |
| 254 | had_speech = true; |
| 255 | silence_samples = 0; |
| 256 | } else if had_speech { |
| 257 | silence_samples += chunk.len() as u32; |
| 258 | } |
| 259 | |
| 260 | if had_speech { |
| 261 | all_samples.extend_from_slice(&chunk); |
| 262 | } |
| 263 | |
| 264 | if start.elapsed() > max_duration { |
| 265 | let _ = child.kill(); |
| 266 | break; |
| 267 | } |
| 268 | if had_speech && silence_samples >= silence_duration_samples { |
| 269 | let _ = child.kill(); |
| 270 | break; |
| 271 | } |
| 272 | } |
| 273 | Err(e) if e.kind() == std::io::ErrorKind::UnexpectedEof => break, |
| 274 | Err(_) => { |
| 275 | let _ = child.kill(); |
| 276 | break; |
| 277 | } |
| 278 | } |
| 279 | } |
| 280 | |
| 281 | let _ = child.wait(); |
| 282 | let elapsed = start.elapsed(); |
| 283 | |
| 284 | let min_samples = (MIN_SEGMENT_SECS * 16000.0) as usize; |
| 285 | if all_samples.len() < min_samples { |
| 286 | return None; |
| 287 | } |
| 288 | |
| 289 | Some((all_samples, elapsed)) |
| 290 | } |
| 291 | |
| 292 | // --- Auto-send suffix ------------------------------------------------------ |
| 293 | |
| 294 | /// Trailing phrases that mean "submit this" — the human-readable form of |
| 295 | /// `SEND_SUFFIX_RE`; keep in sync with the regex when either changes. |
| 296 | pub const SEND_PHRASES: &[&str] = &["send it", "发送", "發送"]; |
| 297 | |
| 298 | /// Matches an explicit send instruction at the end of transcribed text: |
| 299 | /// "send it" (any spacing/case) or 发送/發送, with trailing punctuation. |
| 300 | static SEND_SUFFIX_RE: LazyLock<Regex> = LazyLock::new(|| { |
| 301 | Regex::new(r"(?i)(?:^|[\s,，.。!！?？]+)(?:send\s*it|发送|發送)[\s.。!！?？]*$").unwrap() |
| 302 | }); |
| 303 | |
| 304 | /// Split a transcript into the message remainder and whether it ended with an |
| 305 | /// explicit send instruction. `"ship the fix, send it"` → `("ship the fix", true)`. |
| 306 | fn split_send_suffix(text: &str) -> (&str, bool) { |
| 307 | match SEND_SUFFIX_RE.find(text) { |
| 308 | Some(found) => (text[..found.start()].trim(), true), |
| 309 | None => (text.trim(), false), |
| 310 | } |
| 311 | } |
| 312 | |
| 313 | // --- Transcription --------------------------------------------------------- |
| 314 | |
| 315 | fn base64_encode(data: &[u8]) -> String { |
| 316 | use base64::Engine; |
| 317 | base64::engine::general_purpose::STANDARD.encode(data) |
| 318 | } |
| 319 | |
| 320 | fn chat_completions_url(base_url: &str) -> String { |
| 321 | format!("{}/chat/completions", base_url.trim_end_matches('/')) |
| 322 | } |
| 323 | |
| 324 | async fn post_chat_completions( |
| 325 | api_key: &str, |
| 326 | base_url: &str, |
| 327 | mut body: serde_json::Value, |
| 328 | openrouter_vendor: Option<&str>, |
| 329 | ) -> Result<serde_json::Value, String> { |
| 330 | crate::client::apply_openrouter_vendor(&mut body, openrouter_vendor); |
| 331 | let _inference = crate::client::acquire_remote_control_inference_participant().await; |
| 332 | let client = crate::tls::reqwest_client(); |
| 333 | let resp = client |
| 334 | .post(chat_completions_url(base_url)) |
| 335 | .header("Content-Type", "application/json") |
| 336 | .header("Authorization", format!("Bearer {api_key}")) |
| 337 | .timeout(Duration::from_secs(30)) |
| 338 | .json(&body) |
| 339 | .send() |
| 340 | .await |
| 341 | .map_err(|e| format!("request failed: {e}"))?; |
| 342 | |
| 343 | if !resp.status().is_success() { |
| 344 | return Err(format!("API returned status {}", resp.status())); |
| 345 | } |
| 346 | |
| 347 | resp.json() |
| 348 | .await |
| 349 | .map_err(|e| format!("failed to parse response: {e}")) |
| 350 | } |
| 351 | |
| 352 | /// Send audio to the provider's API for plain transcription. |
| 353 | /// |
| 354 | /// Uses the chat completions endpoint with `input_audio` content blocks. |
| 355 | async fn transcribe( |
| 356 | api_key: &str, |
| 357 | base_url: &str, |
| 358 | audio_samples: &[i16], |
| 359 | openrouter_vendor: Option<&str>, |
| 360 | ) -> Result<String, String> { |
| 361 | transcribe_with_model( |
| 362 | api_key, |
| 363 | base_url, |
| 364 | audio_samples, |
| 365 | ASR_MODEL, |
| 366 | openrouter_vendor, |
| 367 | ) |
| 368 | .await |
| 369 | } |
| 370 | |
| 371 | async fn transcribe_with_model( |
| 372 | api_key: &str, |
| 373 | base_url: &str, |
| 374 | audio_samples: &[i16], |
| 375 | model: &str, |
| 376 | openrouter_vendor: Option<&str>, |
| 377 | ) -> Result<String, String> { |
| 378 | let wav = encode_wav(audio_samples); |
| 379 | let data_url = format!("data:audio/wav;base64,{}", base64_encode(&wav)); |
| 380 | |
| 381 | let body = serde_json::json!({ |
| 382 | "model": model, |
| 383 | "messages": [ |
| 384 | { |
| 385 | "role": "user", |
| 386 | "content": [ |
| 387 | { |
| 388 | "type": "input_audio", |
| 389 | "input_audio": { |
| 390 | "data": data_url |
| 391 | } |
| 392 | } |
| 393 | ] |
| 394 | } |
| 395 | ], |
| 396 | "asr_options": { |
| 397 | "language": "auto" |
| 398 | } |
| 399 | }); |
| 400 | |
| 401 | let data = post_chat_completions(api_key, base_url, body, openrouter_vendor).await?; |
| 402 | data["choices"][0]["message"]["content"] |
| 403 | .as_str() |
| 404 | .map(|s| s.trim().to_string()) |
| 405 | .ok_or_else(|| "no transcription in response".to_string()) |
| 406 | } |
| 407 | |
| 408 | /// Process audio through the voice-control pipeline: AI-assisted dictation |
| 409 | /// that sees the current composer text, mirroring MiMo Code's |
| 410 | /// `processVoiceControl`. Used when `/voice-control` is enabled. |
| 411 | async fn process_voice_control( |
| 412 | api_key: &str, |
| 413 | base_url: &str, |
| 414 | audio_samples: &[i16], |
| 415 | current_text: &str, |
| 416 | openrouter_vendor: Option<&str>, |
| 417 | ) -> Result<String, String> { |
| 418 | let wav = encode_wav(audio_samples); |
| 419 | let data_url = format!("data:audio/wav;base64,{}", base64_encode(&wav)); |
| 420 | |
| 421 | let user_context = serde_json::json!({ |
| 422 | "current_text": current_text, |
| 423 | "cursor": "end", |
| 424 | }); |
| 425 | |
| 426 | let body = serde_json::json!({ |
| 427 | "model": VOICE_CONTROL_MODEL, |
| 428 | "messages": [ |
| 429 | { |
| 430 | "role": "system", |
| 431 | "content": "You are a voice input assistant. Transcribe the user's speech. Output JSON: {\"text\": \"transcribed text\"}." |
| 432 | }, |
| 433 | { |
| 434 | "role": "user", |
| 435 | "content": [ |
| 436 | { "type": "text", "text": user_context.to_string() }, |
| 437 | { "type": "input_audio", "input_audio": { "data": data_url } } |
| 438 | ] |
| 439 | } |
| 440 | ], |
| 441 | "response_format": { "type": "json_object" } |
| 442 | }); |
| 443 | |
| 444 | let data = post_chat_completions(api_key, base_url, body, openrouter_vendor).await?; |
| 445 | let content = data["choices"][0]["message"]["content"] |
| 446 | .as_str() |
| 447 | .ok_or_else(|| "no response content".to_string())?; |
| 448 | |
| 449 | let parsed: serde_json::Value = serde_json::from_str(content) |
| 450 | .map_err(|e| format!("failed to parse voice control JSON: {e}"))?; |
| 451 | |
| 452 | parsed["text"] |
| 453 | .as_str() |
| 454 | .map(|s| s.to_string()) |
| 455 | .ok_or_else(|| "no text field in voice control response".to_string()) |
| 456 | } |
| 457 | |
| 458 | // --- Capture orchestration (UI event loop) --------------------------------- |
| 459 | |
| 460 | /// What the UI should do with a finished capture. |
| 461 | #[derive(Debug, Clone, PartialEq, Eq)] |
| 462 | pub enum VoiceCaptureOutcome { |
| 463 | /// Insert the transcribed text into the composer at the cursor. |
| 464 | Insert(String), |
| 465 | /// Submit this text as a message (auto-send). |
| 466 | Send(String), |
| 467 | } |
| 468 | |
| 469 | /// Detect best free ASR for this host — local whisper > Groq free > provider fallback. |
| 470 | /// Works on macOS (brew install whisper-cpp), Windows (whisper.cpp binary), |
| 471 | /// Linux (apt), and HarmonyOS (falls back to cloud). |
| 472 | fn detect_free_asr() -> &'static str { |
| 473 | for bin in LOCAL_WHISPER_BINS { |
| 474 | if Command::new(bin) |
| 475 | .arg("--help") |
| 476 | .stdin(Stdio::null()) |
| 477 | .stdout(Stdio::null()) |
| 478 | .stderr(Stdio::null()) |
| 479 | .spawn() |
| 480 | .is_ok() |
| 481 | { |
| 482 | return "local-whisper"; |
| 483 | } |
| 484 | } |
| 485 | if std::env::var("GROQ_API_KEY").is_ok_and(|v| !v.trim().is_empty()) { |
| 486 | return "groq"; |
| 487 | } |
| 488 | "provider" |
| 489 | } |
| 490 | |
| 491 | /// Transcribe via local whisper.cpp (free, offline, cross-platform). |
| 492 | /// |
| 493 | /// The whole body is synchronous — temp-file I/O plus `Command::output()`, |
| 494 | /// which blocks for the entire subprocess run — so it runs on the blocking |
| 495 | /// pool rather than a Tokio worker (blocking-call convention, #6149). |
| 496 | async fn transcribe_local_whisper(audio_samples: &[i16]) -> Result<String, String> { |
| 497 | let wav = encode_wav(audio_samples); |
| 498 | tokio::task::spawn_blocking(move || transcribe_local_whisper_blocking(&wav)) |
| 499 | .await |
| 500 | .map_err(|e| e.to_string())? |
| 501 | } |
| 502 | |
| 503 | fn transcribe_local_whisper_blocking(wav: &[u8]) -> Result<String, String> { |
| 504 | let tmp = std::env::temp_dir().join(format!("cw-voice-{}.wav", std::process::id())); |
| 505 | std::fs::write(&tmp, wav).map_err(|e| e.to_string())?; |
| 506 | // Try each local binary until one succeeds; whisper.cpp outputs to stdout or file. |
| 507 | for bin in LOCAL_WHISPER_BINS { |
| 508 | let output = Command::new(bin) |
| 509 | .arg(tmp.to_string_lossy().as_ref()) |
| 510 | .arg("--model") |
| 511 | .arg("tiny") |
| 512 | .arg("--language") |
| 513 | .arg("auto") |
| 514 | .arg("--output-txt") |
| 515 | .output(); |
| 516 | if let Ok(out) = output |
| 517 | && out.status.success() |
| 518 | { |
| 519 | let txt = String::from_utf8_lossy(&out.stdout).trim().to_string(); |
| 520 | let _ = std::fs::remove_file(&tmp); |
| 521 | if !txt.is_empty() { |
| 522 | return Ok(txt); |
| 523 | } |
| 524 | // Some builds write to .txt sidecar |
| 525 | let sidecar = tmp.with_extension("txt"); |
| 526 | if let Ok(s) = std::fs::read_to_string(&sidecar) { |
| 527 | let _ = std::fs::remove_file(&sidecar); |
| 528 | let _ = std::fs::remove_file(&tmp); |
| 529 | if !s.trim().is_empty() { |
| 530 | return Ok(s.trim().to_string()); |
| 531 | } |
| 532 | } |
| 533 | } |
| 534 | } |
| 535 | let _ = std::fs::remove_file(&tmp); |
| 536 | Err("local whisper not available".into()) |
| 537 | } |
| 538 | |
| 539 | /// Transcribe via Groq Whisper large-v3-turbo (free tier, ~$0.04/hr, fast). |
| 540 | /// Groq is NOT a full CodeWhale provider yet — this is a direct ASR call |
| 541 | /// using `GROQ_API_KEY` only (no provider setup needed). Uses the same |
| 542 | /// chat-completions `input_audio` path as Xiaomi so no `multipart` feature. |
| 543 | async fn transcribe_groq(audio_samples: &[i16]) -> Result<String, String> { |
| 544 | let api_key = std::env::var("GROQ_API_KEY").map_err(|_| "GROQ_API_KEY not set".to_string())?; |
| 545 | let base_url = "https://api.groq.com/openai/v1"; |
| 546 | transcribe_with_model(&api_key, base_url, audio_samples, GROQ_ASR_MODEL, None).await |
| 547 | } |
| 548 | |
| 549 | /// Perform a complete record + transcribe cycle with live interim display. |
| 550 | /// |
| 551 | /// Runs in the UI event loop (see [`AppAction::VoiceCapture`]) so provider |
| 552 | /// credentials come from the live [`Config`] rather than state cached on |
| 553 | /// [`App`]. Recording happens on a blocking thread; transcription uses the |
| 554 | /// shared async HTTP client. Every failure path returns a localized message |
| 555 | /// so callers can surface it as a status line. |
| 556 | /// Resolve ASR model/provider preference. |
| 557 | /// Priority: explicit config `voice.asr_model` > env `CODEWHALE_ASR_MODEL` > auto-detect (local-whisper > groq > xiaomi). |
| 558 | fn resolve_asr_choice(_config: &Config) -> (String, String) { |
| 559 | // Check explicit env override first (free, cross-platform) |
| 560 | if let Ok(m) = std::env::var("CODEWHALE_ASR_MODEL") { |
| 561 | let m = m.trim().to_ascii_lowercase(); |
| 562 | if m.contains("groq") || m.contains("whisper") { |
| 563 | return ("groq".into(), GROQ_ASR_MODEL.into()); |
| 564 | } |
| 565 | if m.contains("local") || m.contains("whisper.cpp") { |
| 566 | return ("local-whisper".into(), "tiny".into()); |
| 567 | } |
| 568 | if m.contains("mimo") || m.contains("xiaomi") { |
| 569 | return ("provider".into(), ASR_MODEL.into()); |
| 570 | } |
| 571 | } |
| 572 | // Auto-detect best free: local whisper (offline, no key) > Groq free tier > Xiaomi ASR (needs key) |
| 573 | let free = detect_free_asr(); |
| 574 | match free { |
| 575 | "local-whisper" => ("local-whisper".into(), "tiny".into()), |
| 576 | "groq" => ("groq".into(), GROQ_ASR_MODEL.into()), |
| 577 | _ => ("provider".into(), ASR_MODEL.into()), |
| 578 | } |
| 579 | } |
| 580 | |
| 581 | pub async fn capture_and_transcribe( |
| 582 | app: &mut App, |
| 583 | config: &Config, |
| 584 | ) -> Result<VoiceCaptureOutcome, String> { |
| 585 | let locale = app.ui_locale; |
| 586 | |
| 587 | if !is_available() { |
| 588 | return Err(tr(locale, MessageId::VoiceErrNoRecorder).to_string()); |
| 589 | } |
| 590 | let api_key = config |
| 591 | .active_route_api_key() |
| 592 | .map_err(|_| tr(locale, MessageId::VoiceErrNoAuth).to_string())?; |
| 593 | let base_url = config.active_route_base_url(); |
| 594 | let openrouter_vendor = config |
| 595 | .openrouter_vendor() |
| 596 | .map_err(|error| error.to_string())?; |
| 597 | |
| 598 | // Spark-style: show "● Recording (⌥V to finish)" + live interim in composer. |
| 599 | let original_input = app.composer.input.clone(); |
| 600 | let original_cursor = app.composer.cursor_position; |
| 601 | app.status_message = Some("● Recording (⌥V to finish) · speak naturally".to_string()); |
| 602 | |
| 603 | // Streaming interim: poll every 700ms and show partial transcript like Grok Build's |
| 604 | // VoiceEvent::Interim → VoiceState::Recording{interim}. We re-transcribe the |
| 605 | // growing buffer (local-whisper is cheap; Groq is ~300ms; provider falls back). |
| 606 | let (asr_kind, _asr_model) = resolve_asr_choice(config); |
| 607 | let interim_enabled = true; // always show partials — feels alive like Spark |
| 608 | |
| 609 | // Spawn recorder on blocking thread with a shared buffer for interim polling. |
| 610 | let shared_buf: std::sync::Arc<parking_lot::Mutex<Vec<i16>>> = |
| 611 | std::sync::Arc::new(parking_lot::Mutex::new(Vec::new())); |
| 612 | let shared_done = std::sync::Arc::new(std::sync::atomic::AtomicBool::new(false)); |
| 613 | let shared_buf_clone = std::sync::Arc::clone(&shared_buf); |
| 614 | let shared_done_clone = std::sync::Arc::clone(&shared_done); |
| 615 | let recorder_handle = tokio::task::spawn_blocking(move || { |
| 616 | // Bridge to existing record_audio but copy into shared buffer incrementally. |
| 617 | // For now we reuse the blocking recorder and then publish; interim will |
| 618 | // poll the final buffer. A true streaming recorder (cpal/pw-record) is |
| 619 | // the next step — see grokbuild's xai-grok-voice::audio for the subprocess |
| 620 | // isolation pattern we should mirror. |
| 621 | let result = record_audio(); |
| 622 | if let Some((samples, dur)) = result { |
| 623 | *shared_buf_clone.lock() = samples.clone(); |
| 624 | shared_done_clone.store(true, std::sync::atomic::Ordering::SeqCst); |
| 625 | Some((samples, dur)) |
| 626 | } else { |
| 627 | shared_done_clone.store(true, std::sync::atomic::Ordering::SeqCst); |
| 628 | None |
| 629 | } |
| 630 | }); |
| 631 | |
| 632 | // Interim polling loop — updates composer with "original + interim ▍" so text |
| 633 | // appears as you talk, just like Spark's live transcript. |
| 634 | let mut last_interim = String::new(); |
| 635 | let mut ticks: u32 = 0; |
| 636 | loop { |
| 637 | tokio::time::sleep(Duration::from_millis(700)).await; |
| 638 | ticks += 1; |
| 639 | if shared_done.load(std::sync::atomic::Ordering::SeqCst) { |
| 640 | break; |
| 641 | } |
| 642 | if !interim_enabled || ticks < 2 { |
| 643 | continue; // let a little audio accumulate before first interim |
| 644 | } |
| 645 | let snapshot = { shared_buf.lock().clone() }; |
| 646 | if snapshot.len() < 8000 { |
| 647 | // <0.5s of audio — not enough for meaningful ASR |
| 648 | continue; |
| 649 | } |
| 650 | // Try cheapest free ASR for interim; don't fail the whole capture on interim error. |
| 651 | let interim = match asr_kind.as_str() { |
| 652 | "local-whisper" => transcribe_local_whisper(&snapshot) |
| 653 | .await |
| 654 | .unwrap_or_default(), |
| 655 | "groq" => transcribe_groq(&snapshot).await.unwrap_or_default(), |
| 656 | _ => { |
| 657 | // For provider ASR, reuse the same endpoint but don't block on interim if no key. |
| 658 | if let Ok(key) = config |
| 659 | .active_route_api_key() |
| 660 | .map(|k: String| k) |
| 661 | .map_err(|_| String::new()) |
| 662 | { |
| 663 | let url = config.active_route_base_url(); |
| 664 | transcribe(&key, &url, &snapshot, openrouter_vendor.as_deref()) |
| 665 | .await |
| 666 | .unwrap_or_default() |
| 667 | } else { |
| 668 | String::new() |
| 669 | } |
| 670 | } |
| 671 | }; |
| 672 | let trimmed = interim.trim(); |
| 673 | if !trimmed.is_empty() && trimmed != last_interim { |
| 674 | last_interim = trimmed.to_string(); |
| 675 | // Show interim inline — preserve cursor at original position, append interim with a block cursor |
| 676 | let display = if original_input.trim().is_empty() { |
| 677 | format!("{trimmed} ▍") |
| 678 | } else { |
| 679 | format!("{} {} ▍", original_input.trim_end(), trimmed) |
| 680 | }; |
| 681 | app.composer.input = display; |
| 682 | app.composer.cursor_position = original_cursor; |
| 683 | // Also keep status as Spark does |
| 684 | app.status_message = Some(format!("● Listening — “{trimmed}” (⌥V to finish)")); |
| 685 | } |
| 686 | if ticks > 40 { |
| 687 | break; // safety: ~28s max interim polling |
| 688 | } |
| 689 | } |
| 690 | |
| 691 | let (samples, _duration) = recorder_handle |
| 692 | .await |
| 693 | .ok() |
| 694 | .flatten() |
| 695 | .ok_or_else(|| tr(locale, MessageId::VoiceErrTooShort).to_string())?; |
| 696 | |
| 697 | // Restore composer to original before final insert (interim was preview only) |
| 698 | app.composer.input = original_input.clone(); |
| 699 | app.composer.cursor_position = original_cursor; |
| 700 | app.status_message = Some(tr(locale, MessageId::VoiceProcessing).to_string()); |
| 701 | |
| 702 | let text = match asr_kind.as_str() { |
| 703 | "local-whisper" => match transcribe_local_whisper(&samples).await { |
| 704 | Ok(v) => Ok(v), |
| 705 | Err(_) => transcribe(&api_key, &base_url, &samples, openrouter_vendor.as_deref()).await, |
| 706 | }, |
| 707 | "groq" => match transcribe_groq(&samples).await { |
| 708 | Ok(v) => Ok(v), |
| 709 | Err(_) => transcribe(&api_key, &base_url, &samples, openrouter_vendor.as_deref()).await, |
| 710 | }, |
| 711 | _ => { |
| 712 | if app.voice_control_enabled { |
| 713 | process_voice_control( |
| 714 | &api_key, |
| 715 | &base_url, |
| 716 | &samples, |
| 717 | &original_input, |
| 718 | openrouter_vendor.as_deref(), |
| 719 | ) |
| 720 | .await |
| 721 | } else { |
| 722 | transcribe(&api_key, &base_url, &samples, openrouter_vendor.as_deref()).await |
| 723 | } |
| 724 | } |
| 725 | } |
| 726 | .map_err(|e| format!("{}: {e}", tr(locale, MessageId::VoiceErrNetwork)))?; |
| 727 | |
| 728 | let clean = text.trim(); |
| 729 | if app.voice_send_enabled { |
| 730 | let (remainder, wants_send) = split_send_suffix(clean); |
| 731 | if wants_send { |
| 732 | // A bare "send it" submits whatever is already in the composer. |
| 733 | let outgoing = if remainder.is_empty() { |
| 734 | let existing = app.composer.input.trim().to_string(); |
| 735 | if !existing.is_empty() { |
| 736 | app.clear_input(); |
| 737 | } |
| 738 | existing |
| 739 | } else { |
| 740 | remainder.to_string() |
| 741 | }; |
| 742 | if outgoing.is_empty() { |
| 743 | return Err(tr(locale, MessageId::VoiceErrEmptySend).to_string()); |
| 744 | } |
| 745 | return Ok(VoiceCaptureOutcome::Send(outgoing)); |
| 746 | } |
| 747 | } |
| 748 | if clean.is_empty() { |
| 749 | return Err(tr(locale, MessageId::VoiceErrEmptySend).to_string()); |
| 750 | } |
| 751 | Ok(VoiceCaptureOutcome::Insert(clean.to_string())) |
| 752 | } |
| 753 | |
| 754 | // --- Headless capture (HTTP/native-client path) ---------------------------- |
| 755 | |
| 756 | /// What a headless dictation should do with the finished transcript. |
| 757 | #[derive(Debug, Clone)] |
| 758 | pub enum DictateMode { |
| 759 | /// Transcribe and return the text for insertion into the composer. |
| 760 | Insert, |
| 761 | /// Transcribe, then apply the "send it" / 发送 suffix contract. The |
| 762 | /// outcome's `send` flag tells the client to submit; a bare send |
| 763 | /// instruction yields empty `text` so the client submits its own draft. |
| 764 | Send, |
| 765 | /// AI-assisted dictation that sees the client's composer text — the |
| 766 | /// `/voice-control` pipeline. Only provider ASR can see context; free |
| 767 | /// ASR kinds degrade to plain transcription with `assisted: false`. |
| 768 | Control(String), |
| 769 | } |
| 770 | |
| 771 | /// Machine-readable failure for the headless path so HTTP clients can |
| 772 | /// localize by `reason` rather than parsing message text. |
| 773 | #[derive(Debug)] |
| 774 | pub enum DictateError { |
| 775 | /// No supported recorder binary on this host. |
| 776 | NoRecorder, |
| 777 | /// Recording produced no usable speech segment. |
| 778 | NoSpeech, |
| 779 | /// The selected/fallback ASR needs a provider key that isn't configured. |
| 780 | NoProviderAuth, |
| 781 | /// ASR request or transcription failed. |
| 782 | Transcription(String), |
| 783 | } |
| 784 | |
| 785 | impl DictateError { |
| 786 | pub fn reason(&self) -> &'static str { |
| 787 | match self { |
| 788 | Self::NoRecorder => "no_recorder", |
| 789 | Self::NoSpeech => "no_speech", |
| 790 | Self::NoProviderAuth => "no_provider_auth", |
| 791 | Self::Transcription(_) => "transcription_failed", |
| 792 | } |
| 793 | } |
| 794 | } |
| 795 | |
| 796 | impl std::fmt::Display for DictateError { |
| 797 | fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { |
| 798 | match self { |
| 799 | Self::NoRecorder => write!(f, "no supported voice recorder on this host"), |
| 800 | Self::NoSpeech => write!(f, "no speech detected"), |
| 801 | Self::NoProviderAuth => { |
| 802 | write!(f, "provider ASR requires a configured API key") |
| 803 | } |
| 804 | Self::Transcription(e) => write!(f, "{e}"), |
| 805 | } |
| 806 | } |
| 807 | } |
| 808 | |
| 809 | /// Result of one headless record→transcribe cycle. |
| 810 | #[derive(Debug)] |
| 811 | pub struct DictationOutcome { |
| 812 | /// Final transcript (send suffix already stripped for `Send` mode). |
| 813 | pub text: String, |
| 814 | /// `Send` mode only: the transcript ended with an explicit send phrase. |
| 815 | pub send: bool, |
| 816 | /// `Control` mode only: the composer context reached the model. False |
| 817 | /// when a free ASR kind handled the audio and never saw the context. |
| 818 | pub assisted: bool, |
| 819 | /// Which ASR backend was selected for this capture. |
| 820 | pub asr_kind: String, |
| 821 | pub asr_model: String, |
| 822 | } |
| 823 | |
| 824 | /// Detected recorder binary name, if any (`"sox"`, `"arecord"`, `"rec"`). |
| 825 | pub fn recorder_command() -> Option<&'static str> { |
| 826 | detect_recorder().map(|r| r.cmd) |
| 827 | } |
| 828 | |
| 829 | /// Resolved ASR selection (`kind`, `model`) for capability reporting. |
| 830 | pub fn asr_choice(config: &Config) -> (String, String) { |
| 831 | resolve_asr_choice(config) |
| 832 | } |
| 833 | |
| 834 | /// One record→transcribe cycle with no UI surface: the HTTP/native-client |
| 835 | /// equivalent of [`capture_and_transcribe`]. Recording runs on a blocking |
| 836 | /// thread; transcription follows the same ASR dispatch as the TUI — |
| 837 | /// explicit `CODEWHALE_ASR_MODEL` > local whisper > Groq > provider — |
| 838 | /// but resolves the provider key lazily so free ASR kinds work without |
| 839 | /// provider auth. |
| 840 | pub async fn dictate_once( |
| 841 | config: &Config, |
| 842 | mode: DictateMode, |
| 843 | ) -> Result<DictationOutcome, DictateError> { |
| 844 | if !is_available() { |
| 845 | return Err(DictateError::NoRecorder); |
| 846 | } |
| 847 | let (samples, _duration) = tokio::task::spawn_blocking(record_audio) |
| 848 | .await |
| 849 | .ok() |
| 850 | .flatten() |
| 851 | .ok_or(DictateError::NoSpeech)?; |
| 852 | |
| 853 | let (asr_kind, asr_model) = resolve_asr_choice(config); |
| 854 | let base_url = config.active_route_base_url(); |
| 855 | let openrouter_vendor = config |
| 856 | .openrouter_vendor() |
| 857 | .map_err(|e| DictateError::Transcription(e.to_string()))?; |
| 858 | let provider_key = || { |
| 859 | config |
| 860 | .active_route_api_key() |
| 861 | .map_err(|_| DictateError::NoProviderAuth) |
| 862 | }; |
| 863 | |
| 864 | let mut assisted = false; |
| 865 | let text = match asr_kind.as_str() { |
| 866 | "local-whisper" => match transcribe_local_whisper(&samples).await { |
| 867 | Ok(v) => v, |
| 868 | Err(_) => transcribe( |
| 869 | &provider_key()?, |
| 870 | &base_url, |
| 871 | &samples, |
| 872 | openrouter_vendor.as_deref(), |
| 873 | ) |
| 874 | .await |
| 875 | .map_err(DictateError::Transcription)?, |
| 876 | }, |
| 877 | "groq" => match transcribe_groq(&samples).await { |
| 878 | Ok(v) => v, |
| 879 | Err(_) => transcribe( |
| 880 | &provider_key()?, |
| 881 | &base_url, |
| 882 | &samples, |
| 883 | openrouter_vendor.as_deref(), |
| 884 | ) |
| 885 | .await |
| 886 | .map_err(DictateError::Transcription)?, |
| 887 | }, |
| 888 | _ => { |
| 889 | let api_key = provider_key()?; |
| 890 | match &mode { |
| 891 | DictateMode::Control(composer) => { |
| 892 | assisted = true; |
| 893 | process_voice_control( |
| 894 | &api_key, |
| 895 | &base_url, |
| 896 | &samples, |
| 897 | composer, |
| 898 | openrouter_vendor.as_deref(), |
| 899 | ) |
| 900 | .await |
| 901 | .map_err(DictateError::Transcription)? |
| 902 | } |
| 903 | _ => transcribe(&api_key, &base_url, &samples, openrouter_vendor.as_deref()) |
| 904 | .await |
| 905 | .map_err(DictateError::Transcription)?, |
| 906 | } |
| 907 | } |
| 908 | }; |
| 909 | |
| 910 | let clean = text.trim().to_string(); |
| 911 | let (text, send) = match mode { |
| 912 | DictateMode::Send => { |
| 913 | let (remainder, wants_send) = split_send_suffix(&clean); |
| 914 | (remainder.to_string(), wants_send) |
| 915 | } |
| 916 | _ => (clean, false), |
| 917 | }; |
| 918 | Ok(DictationOutcome { |
| 919 | text, |
| 920 | send, |
| 921 | assisted, |
| 922 | asr_kind, |
| 923 | asr_model, |
| 924 | }) |
| 925 | } |
| 926 | |
| 927 | // --- Command handlers ------------------------------------------------------ |
| 928 | |
| 929 | /// Handle the `/voice` command: toggle voice input. Toggling on requests a |
| 930 | /// one-shot recording + transcription via [`AppAction::VoiceCapture`]. |
| 931 | pub fn voice(app: &mut App) -> CommandResult { |
| 932 | let locale = app.ui_locale; |
| 933 | |
| 934 | if app.voice_enabled { |
| 935 | app.voice_enabled = false; |
| 936 | return CommandResult::message(tr(locale, MessageId::VoiceDisabled)); |
| 937 | } |
| 938 | if !is_available() { |
| 939 | return CommandResult::error(tr(locale, MessageId::VoiceErrNoRecorder)); |
| 940 | } |
| 941 | app.voice_enabled = true; |
| 942 | CommandResult::with_message_and_action( |
| 943 | tr(locale, MessageId::VoiceEnabled), |
| 944 | AppAction::VoiceCapture, |
| 945 | ) |
| 946 | } |
| 947 | |
| 948 | /// Handle the `/voice-send` command: toggle auto-send after transcription. |
| 949 | pub fn voice_send(app: &mut App) -> CommandResult { |
| 950 | let locale = app.ui_locale; |
| 951 | app.voice_send_enabled = !app.voice_send_enabled; |
| 952 | |
| 953 | let msg = if app.voice_send_enabled { |
| 954 | tr(locale, MessageId::VoiceSendEnabled) |
| 955 | } else { |
| 956 | tr(locale, MessageId::VoiceSendDisabled) |
| 957 | }; |
| 958 | CommandResult::message(msg) |
| 959 | } |
| 960 | |
| 961 | /// Handle the `/voice-control` command: toggle AI-assisted dictation. |
| 962 | pub fn voice_control(app: &mut App) -> CommandResult { |
| 963 | let locale = app.ui_locale; |
| 964 | app.voice_control_enabled = !app.voice_control_enabled; |
| 965 | |
| 966 | let msg = if app.voice_control_enabled { |
| 967 | tr(locale, MessageId::VoiceControlEnabled) |
| 968 | } else { |
| 969 | tr(locale, MessageId::VoiceControlDisabled) |
| 970 | }; |
| 971 | CommandResult::message(msg) |
| 972 | } |
| 973 | |
| 974 | #[cfg(test)] |
| 975 | mod tests { |
| 976 | use super::*; |
| 977 | |
| 978 | #[tokio::test] |
| 979 | async fn voice_requests_preserve_openrouter_vendor_pin() { |
| 980 | use wiremock::matchers::{method, path}; |
| 981 | use wiremock::{Mock, MockServer, ResponseTemplate}; |
| 982 | |
| 983 | let server = MockServer::start().await; |
| 984 | Mock::given(method("POST")) |
| 985 | .and(path("/v1/chat/completions")) |
| 986 | .respond_with(ResponseTemplate::new(200).set_body_json(serde_json::json!({ |
| 987 | "choices": [{ "message": { "content": "{\"text\":\"hello\"}" } }] |
| 988 | }))) |
| 989 | .expect(3) |
| 990 | .mount(&server) |
| 991 | .await; |
| 992 | let base_url = format!("{}/v1", server.uri()); |
| 993 | transcribe( |
| 994 | "fixture-key", |
| 995 | &base_url, |
| 996 | &[0; 16], |
| 997 | Some("chutes/region-fixture"), |
| 998 | ) |
| 999 | .await |
| 1000 | .unwrap(); |
| 1001 | process_voice_control( |
| 1002 | "fixture-key", |
| 1003 | &base_url, |
| 1004 | &[0; 16], |
| 1005 | "existing text", |
| 1006 | Some("chutes/region-fixture"), |
| 1007 | ) |
| 1008 | .await |
| 1009 | .unwrap(); |
| 1010 | transcribe_with_model("fixture-key", &base_url, &[0; 16], GROQ_ASR_MODEL, None) |
| 1011 | .await |
| 1012 | .unwrap(); |
| 1013 | |
| 1014 | let requests = server.received_requests().await.unwrap(); |
| 1015 | assert_eq!(requests.len(), 3); |
| 1016 | for request in &requests[..2] { |
| 1017 | let body: serde_json::Value = serde_json::from_slice(&request.body).unwrap(); |
| 1018 | assert_eq!( |
| 1019 | body["provider"], |
| 1020 | serde_json::json!({"order": ["chutes/region-fixture"], "allow_fallbacks": false}) |
| 1021 | ); |
| 1022 | } |
| 1023 | let independent: serde_json::Value = serde_json::from_slice(&requests[2].body).unwrap(); |
| 1024 | assert!(independent.get("provider").is_none()); |
| 1025 | } |
| 1026 | |
| 1027 | #[test] |
| 1028 | fn wav_encoding_produces_valid_header() { |
| 1029 | let samples = vec![0i16; 16000]; // 1 second of silence |
| 1030 | let wav = encode_wav(&samples); |
| 1031 | assert_eq!(&wav[0..4], b"RIFF"); |
| 1032 | assert_eq!(&wav[8..12], b"WAVE"); |
| 1033 | assert_eq!(&wav[12..16], b"fmt "); |
| 1034 | // data size = 16000 * 2 = 32000 |
| 1035 | assert_eq!(&wav[4..8], &(36 + 32000u32).to_le_bytes()); |
| 1036 | } |
| 1037 | |
| 1038 | #[test] |
| 1039 | fn wav_encoding_empty_is_minimal() { |
| 1040 | let wav = encode_wav(&[]); |
| 1041 | assert_eq!(wav.len(), 44); |
| 1042 | assert_eq!(&wav[4..8], &36u32.to_le_bytes()); |
| 1043 | } |
| 1044 | |
| 1045 | #[test] |
| 1046 | fn send_suffix_detected_and_stripped() { |
| 1047 | assert_eq!(split_send_suffix("send it"), ("", true)); |
| 1048 | assert_eq!(split_send_suffix("Send It!"), ("", true)); |
| 1049 | assert_eq!(split_send_suffix("发送"), ("", true)); |
| 1050 | assert_eq!(split_send_suffix("發送。"), ("", true)); |
| 1051 | assert_eq!( |
| 1052 | split_send_suffix("ship the fix, send it"), |
| 1053 | ("ship the fix", true) |
| 1054 | ); |
| 1055 | assert_eq!( |
| 1056 | split_send_suffix("修复这个问题，发送"), |
| 1057 | ("修复这个问题", true) |
| 1058 | ); |
| 1059 | } |
| 1060 | |
| 1061 | #[test] |
| 1062 | fn send_suffix_leaves_plain_text_alone() { |
| 1063 | assert_eq!(split_send_suffix("send it now"), ("send it now", false)); |
| 1064 | assert_eq!( |
| 1065 | split_send_suffix("帮我发送一封邮件"), |
| 1066 | ("帮我发送一封邮件", false) |
| 1067 | ); |
| 1068 | assert_eq!(split_send_suffix("发送邮件"), ("发送邮件", false)); |
| 1069 | assert_eq!( |
| 1070 | split_send_suffix("resend it to the queue"), |
| 1071 | ("resend it to the queue", false) |
| 1072 | ); |
| 1073 | } |
| 1074 | |
| 1075 | #[test] |
| 1076 | fn recorder_detection_does_not_crash() { |
| 1077 | // Just verify the function runs without panicking |
| 1078 | let _ = is_available(); |
| 1079 | } |
| 1080 | } |
| 1081 |