From a21e46cbb9839c5baf6cdca8d84c1415768f86d8 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E5=90=95=E6=9F=8F=E9=9D=92?= Date: Sun, 7 Jun 2026 12:35:18 +0800 Subject: [PATCH] =?UTF-8?q?refactor(coordinator):=20=E6=8A=BD=E5=87=BA=20a?= =?UTF-8?q?sr=5Fsetup=20=E5=AD=90=E6=A8=A1=E5=9D=97?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 机械式行为保持拆分：从 coordinator.rs 搬出 ASR 引擎构建与生命周期 —— 麦克风/凭证 前置校验、本地模型释放调度（Foundry/sherpa/local-qwen3）、QaAsrStart provider 构建、 全局超时常量与各 provider 超时、音频线程驱动的 DeferredAsrBridge。schedule_*_release session 守卫与 DeferredAsrBridge attach/flush 交错原样保留。ActiveAsr 及 QaAsrStart/ DeferredAsrBridge 的跨模块方法提升为 pub(crate)；ActiveAsr/分类器留在父模块。 Co-Authored-By: Claude Opus 4.8 (1M context) --- openless-all/app/src-tauri/src/coordinator.rs | 495 +---------------- .../src-tauri/src/coordinator/asr_setup.rs | 506 ++++++++++++++++++ 2 files changed, 509 insertions(+), 492 deletions(-) create mode 100644 openless-all/app/src-tauri/src/coordinator/asr_setup.rs diff --git a/openless-all/app/src-tauri/src/coordinator.rs b/openless-all/app/src-tauri/src/coordinator.rs index a0b2c86a..a3bed14b 100644 --- a/openless-all/app/src-tauri/src/coordinator.rs +++ b/openless-all/app/src-tauri/src/coordinator.rs @@ -61,6 +61,7 @@ use crate::windows_ime_ipc::ImeSubmitTarget; #[cfg(target_os = "windows")] use crate::windows_ime_session::{PreparedWindowsImeSession, WindowsImeSessionController}; +mod asr_setup; mod capsule; mod dictation; mod hotkey_supervisors; @@ -72,6 +73,7 @@ mod voice_agent_hotkeys; // glob 重导出：让 dictation.rs/qa.rs/resources.rs/impl 里所有 `super::裸名` // 引用继续通过父模块解析（拆分前的 `use super::*` 契约）。 +pub(crate) use asr_setup::*; pub(crate) use capsule::*; pub(crate) use hotkey_supervisors::*; pub(crate) use ime_insertion::*; @@ -96,7 +98,7 @@ use resources::{ }; #[derive(Clone)] -enum ActiveAsr { +pub(crate) enum ActiveAsr { Volcengine(Arc), Whisper(Arc), Mimo(Arc), @@ -1146,401 +1148,6 @@ fn debug_transcript_override_text() -> Option { } } -fn ensure_microphone_permission(_inner: &Arc) -> Result<(), String> { - use crate::permissions::{self, PermissionStatus}; - - #[cfg(target_os = "windows")] - { - if permissions::windows_microphone_access_explicitly_denied() { - return Err("需要麦克风权限，当前状态: Denied".to_string()); - } - return Ok(()); - } - - let status = permissions::check_microphone(); - if matches!( - status, - PermissionStatus::Granted | PermissionStatus::NotApplicable - ) { - return Ok(()); - } - - // 听写路径不抢前台焦点：缺 mic 权限时直接请求系统授权，不再先 show_main_window。 - // 用户在设置页手动点“请求权限”仍走 request_microphone_from_foreground，那是显式操作。 - // 这里若系统不弹框，后续会通过 capsule error 引导用户主动去权限页处理。详见 #166。 - let requested = permissions::request_microphone(); - if matches!( - requested, - PermissionStatus::Granted | PermissionStatus::NotApplicable - ) { - Ok(()) - } else { - Err(format!("需要麦克风权限，当前状态: {requested:?}")) - } -} - -fn ensure_asr_credentials() -> Result<(), String> { - let active_asr = CredentialsVault::get_active_asr(); - - // 本地 Qwen3-ASR 没有"凭据"概念，但需要：(a) macOS 平台 (b) 模型已下载。 - if crate::asr::local::is_local_qwen3(&active_asr) { - #[cfg(not(target_os = "macos"))] - { - return Err("本地 ASR 当前仅支持 macOS（Windows 见 issue #256）".to_string()); - } - #[cfg(target_os = "macos")] - { - return ensure_local_qwen3_model_ready(); - } - } - - if crate::asr::local::foundry::is_foundry_local_whisper(&active_asr) { - #[cfg(not(target_os = "windows"))] - { - return Err("Foundry Local Whisper 当前仅支持 Windows".to_string()); - } - #[cfg(target_os = "windows")] - { - return Ok(()); - } - } - - if crate::asr::local::sherpa::is_sherpa_onnx_local(&active_asr) { - #[cfg(not(target_os = "windows"))] - { - return Err("sherpa-onnx local ASR 当前仅支持 Windows".to_string()); - } - #[cfg(target_os = "windows")] - { - return Ok(()); - } - } - - if is_whisper_compatible_provider(&active_asr) || is_bailian_provider(&active_asr) { - let api_key = CredentialsVault::get(CredentialAccount::AsrApiKey) - .ok() - .flatten() - .unwrap_or_default(); - if api_key.trim().is_empty() { - return Err("请先在设置中填写 ASR 服务商 API Key".to_string()); - } - return Ok(()); - } - - let creds = read_volc_credentials(); - if creds.app_id.trim().is_empty() || creds.access_token.trim().is_empty() { - Err("请先在设置中填写火山引擎 ASR App Key 和 Access Key".to_string()) - } else { - Ok(()) - } -} - -#[cfg(test)] -fn is_keyless_local_asr_provider(id: &str) -> bool { - if crate::asr::local::is_local_qwen3(id) { - return true; - } - #[cfg(target_os = "windows")] - { - crate::asr::local::foundry::is_foundry_local_whisper(id) - || crate::asr::local::sherpa::is_sherpa_onnx_local(id) - } - #[cfg(not(target_os = "windows"))] - { - let _ = id; - false - } -} - -#[cfg(target_os = "macos")] -fn ensure_local_qwen3_model_ready() -> Result<(), String> { - let prefs = || -> Result { - // 这里没法拿到 inner，直接读 preferences.json 即可（Coordinator 写盘后总是同步的）。 - crate::persistence::PreferencesStore::new() - .map_err(|e| e.to_string()) - .map(|s| s.get()) - }()?; - let model_id = crate::asr::local::ModelId::from_str(&prefs.local_asr_active_model) - .ok_or_else(|| format!("未知的本地模型 id: {}", prefs.local_asr_active_model))?; - if !crate::asr::local::models::is_downloaded(model_id) { - return Err(format!( - "本地模型 {} 未下载完整，请到 设置 → 模型设置 中下载", - model_id.as_str() - )); - } - Ok(()) -} - -/// 一次 dictation 结束后，按 prefs.local_asr_keep_loaded_secs 决定何时释放 -/// 内存里的 Qwen3-ASR 引擎。0 = 立即释放；其它值 = sleep N 秒后看 last_used。 -/// 多次会话叠加多个 sleep 任务，每个独立 check：只要中间又被使用过就跳过释放。 -fn schedule_local_asr_release(inner: &Arc) { - let keep_secs = inner.prefs.get().local_asr_keep_loaded_secs; - let cache = Arc::clone(&inner.local_asr_cache); - if keep_secs == 0 { - cache.release_now(); - return; - } - let dur = std::time::Duration::from_secs(keep_secs as u64); - tauri::async_runtime::spawn(async move { - tokio::time::sleep(dur).await; - cache.release_if_idle(dur); - }); -} - -#[cfg(target_os = "windows")] -fn foundry_local_asr_release_keep_secs(inner: &Arc) -> u32 { - inner.prefs.get().foundry_local_asr_keep_loaded_secs -} - -#[cfg(target_os = "windows")] -#[derive(Clone, Copy)] -enum AsrReleaseSession { - Dictation(SessionId), - Qa(SessionId), -} - -#[cfg(target_os = "windows")] -fn asr_release_session_is_current(inner: &Arc, session: AsrReleaseSession) -> bool { - match session { - AsrReleaseSession::Dictation(session_id) => inner.state.lock().session_id == session_id, - AsrReleaseSession::Qa(session_id) => inner.qa_state.lock().session_id == session_id, - } -} - -#[cfg(target_os = "windows")] -fn schedule_foundry_local_asr_release(inner: &Arc, session: AsrReleaseSession) { - let keep_secs = foundry_local_asr_release_keep_secs(inner); - let runtime = Arc::clone(&inner.foundry_local_runtime); - let inner = Arc::clone(inner); - tauri::async_runtime::spawn(async move { - if keep_secs > 0 { - tokio::time::sleep(std::time::Duration::from_secs(keep_secs as u64)).await; - } - if !asr_release_session_is_current(&inner, session) { - return; - } - if let Err(error) = runtime.release_now().await { - log::warn!("[foundry-asr] scheduled release failed: {error:#}"); - } - }); -} - -#[cfg(target_os = "windows")] -fn sherpa_onnx_release_keep_secs(inner: &Arc) -> u32 { - inner.prefs.get().sherpa_onnx_keep_loaded_secs -} - -/// 与 `schedule_foundry_local_asr_release` 同形：session_id 老旧则不释放， -/// 避免下一轮 session 立即重加载同一个 offline batch 模型。 -#[cfg(target_os = "windows")] -fn schedule_sherpa_onnx_release(inner: &Arc, session: AsrReleaseSession) { - let keep_secs = sherpa_onnx_release_keep_secs(inner); - let runtime = Arc::clone(&inner.sherpa_onnx_runtime); - let inner = Arc::clone(inner); - tauri::async_runtime::spawn(async move { - if keep_secs > 0 { - tokio::time::sleep(std::time::Duration::from_secs(keep_secs as u64)).await; - } - if !asr_release_session_is_current(&inner, session) { - return; - } - if let Err(error) = runtime.release_now().await { - log::warn!("[sherpa-asr] scheduled release failed: {error:#}"); - } - }); -} - -#[cfg(target_os = "macos")] -async fn build_local_qwen3( - inner: &Arc, -) -> anyhow::Result> { - let prefs = inner.prefs.get(); - let model_id = crate::asr::local::ModelId::from_str(&prefs.local_asr_active_model) - .ok_or_else(|| anyhow::anyhow!("未知本地模型 id: {}", prefs.local_asr_active_model))?; - let dir = crate::asr::local::models::model_dir(model_id)?; - let app = inner - .app - .lock() - .clone() - .ok_or_else(|| anyhow::anyhow!("AppHandle 未绑定"))?; - // 走缓存：如果已有同 id 的引擎在内存里就直接复用，避免每次会话都重加载 - // 1.2GB+ 模型。第一次加载阻塞数秒，spawn_blocking 不卡 tokio runtime。 - let cache = Arc::clone(&inner.local_asr_cache); - let mid = model_id.as_str().to_string(); - let engine = tauri::async_runtime::spawn_blocking(move || cache.get_or_load(&mid, &dir)) - .await - .map_err(|e| anyhow::anyhow!("spawn_blocking join failed: {e:#}"))??; - Ok(Arc::new(crate::asr::local::LocalQwenAsr::new(app, engine))) -} - -enum QaAsrStart { - Volcengine { - asr: Arc, - bridge: Arc, - }, - Bailian { - asr: Arc, - bridge: Arc, - }, - Ready { - active: ActiveAsr, - consumer: Arc, - }, -} - -impl QaAsrStart { - fn active_asr(&self) -> ActiveAsr { - match self { - QaAsrStart::Volcengine { asr, .. } => ActiveAsr::Volcengine(Arc::clone(asr)), - QaAsrStart::Bailian { asr, .. } => ActiveAsr::Bailian(Arc::clone(asr)), - QaAsrStart::Ready { active, .. } => active.clone(), - } - } - - fn recorder_consumer(&self) -> Arc { - match self { - QaAsrStart::Volcengine { bridge, .. } => Arc::clone(bridge) as _, - QaAsrStart::Bailian { bridge, .. } => Arc::clone(bridge) as _, - QaAsrStart::Ready { consumer, .. } => Arc::clone(consumer), - } - } - - async fn open_streaming_session(&self) -> Result<(), String> { - match self { - QaAsrStart::Volcengine { asr, bridge } => { - asr.open_session().await.map_err(|e| e.to_string())?; - let target: Arc = Arc::clone(asr) as _; - let flushed = bridge.attach(target); - log::info!("[coord] QA ASR connected; flushed {flushed} deferred audio bytes"); - Ok(()) - } - QaAsrStart::Bailian { asr, bridge } => { - asr.open_session().await.map_err(|e| e.to_string())?; - let target: Arc = Arc::clone(asr) as _; - let flushed = bridge.attach(target); - log::info!( - "[coord] QA Bailian ASR connected; flushed {flushed} deferred audio bytes" - ); - Ok(()) - } - QaAsrStart::Ready { .. } => Ok(()), - } - } -} - -async fn build_qa_asr_start(inner: &Arc, active_asr: &str) -> Result { - #[cfg(target_os = "windows")] - if foundry::is_foundry_local_whisper(active_asr) { - let prefs = inner.prefs.get(); - let model_alias = if foundry::model_alias_is_known(&prefs.foundry_local_asr_model) { - prefs.foundry_local_asr_model.clone() - } else { - foundry::DEFAULT_MODEL_ALIAS.to_string() - }; - let language_hint = prefs.foundry_local_asr_language_hint.trim().to_string(); - let language_hint = if language_hint.is_empty() { - None - } else { - Some(language_hint) - }; - let local = Arc::new(FoundryLocalWhisperAsr::new( - Arc::clone(&inner.foundry_local_runtime), - model_alias, - prefs.foundry_local_runtime_source.clone(), - language_hint, - )); - let active = ActiveAsr::FoundryLocalWhisper(Arc::clone(&local)); - let consumer: Arc = local; - return Ok(QaAsrStart::Ready { active, consumer }); - } - - #[cfg(target_os = "windows")] - if sherpa::is_sherpa_onnx_local(active_asr) { - let prefs = inner.prefs.get(); - let model_alias = if sherpa::model_alias_is_known(&prefs.sherpa_onnx_model) { - prefs.sherpa_onnx_model.clone() - } else { - sherpa::DEFAULT_MODEL_ALIAS.to_string() - }; - let language_hint = prefs.sherpa_onnx_language_hint.trim().to_string(); - let language_hint = if language_hint.is_empty() { - None - } else { - Some(language_hint) - }; - let token_handler = inner.app.lock().clone().map(|app| { - Arc::new(move |piece: String| { - if let Err(error) = app.emit("local-asr-token", piece) { - log::warn!("[sherpa-asr] emit token failed: {error}"); - } - }) as crate::asr::local::sherpa_provider::SherpaTokenHandler - }); - let local = SherpaOnnxAsr::new_for_model( - Arc::clone(&inner.sherpa_onnx_runtime), - model_alias, - language_hint, - token_handler, - ) - .await - .map_err(|e| format!("sherpa-onnx init failed: {e}"))?; - let local = Arc::new(local); - let active = ActiveAsr::SherpaOnnxLocal(Arc::clone(&local)); - let consumer: Arc = local; - return Ok(QaAsrStart::Ready { active, consumer }); - } - - #[cfg(target_os = "macos")] - if crate::asr::local::is_local_qwen3(active_asr) { - let local = build_local_qwen3(inner) - .await - .map_err(|e| format!("local ASR init failed: {e}"))?; - let active = ActiveAsr::Local(Arc::clone(&local)); - let consumer: Arc = local; - return Ok(QaAsrStart::Ready { active, consumer }); - } - - match active_asr_provider_kind(active_asr) { - ActiveAsrProviderKind::Bailian => Ok(QaAsrStart::Bailian { - asr: Arc::new(BailianRealtimeASR::new(read_bailian_credentials())), - bridge: Arc::new(DeferredAsrBridge::new()), - }), - ActiveAsrProviderKind::Mimo => { - let (api_key, base_url, model) = read_mimo_credentials(); - let mimo = Arc::new(MimoBatchASR::new(api_key, base_url, model)); - let active = ActiveAsr::Mimo(Arc::clone(&mimo)); - let consumer: Arc = mimo; - Ok(QaAsrStart::Ready { active, consumer }) - } - ActiveAsrProviderKind::WhisperCompatible => { - let (api_key, base_url, model) = read_whisper_credentials(); - let whisper_prompt = - crate::asr::whisper::build_prompt_from_phrases(&enabled_phrases(inner)); - let whisper = Arc::new( - WhisperBatchASR::new( - api_key, - base_url, - model, - whisper_prompt, - batch_asr_chunk_limit_ms(active_asr), - whisper_supports_verbose_json(active_asr), - ) - .with_request_format(whisper_request_format(active_asr)), - ); - let active = ActiveAsr::Whisper(Arc::clone(&whisper)); - let consumer: Arc = whisper; - Ok(QaAsrStart::Ready { active, consumer }) - } - ActiveAsrProviderKind::Volcengine => Ok(QaAsrStart::Volcengine { - asr: Arc::new(VolcengineStreamingASR::new( - read_volc_credentials(), - enabled_hotwords(inner), - )), - bridge: Arc::new(DeferredAsrBridge::new()), - }), - } -} - // ─────────────────────────── QA session lifecycle ─────────────────────────── /// 划词语音问答会话（issue #118）。 @@ -3045,34 +2652,6 @@ mod tests { } } -/// Coordinator 全局超时保护：防止 ASR await_final_result() 永远挂起。 -/// 设置为 15 秒（比 ASR 的 12 秒 FINAL_RESULT_TIMEOUT 稍长）， -/// 只在 ASR 超时机制失效时作为最后的防线触发。 -const COORDINATOR_GLOBAL_TIMEOUT_SECS: u64 = 15; - -#[cfg(target_os = "windows")] -fn foundry_audio_transcribe_timeout_duration() -> std::time::Duration { - std::time::Duration::from_secs(COORDINATOR_GLOBAL_TIMEOUT_SECS) -} - -/// 本地 Qwen3-ASR 的动态转写超时。固定 15 秒在长录音（≥ 30s）+ 慢机器 -/// （RTF ≈ 0.3–0.5）上必然超时把整段内容丢掉。改用 max(15, ceil(audio_s -/// × 0.6) + 10)：基础保留 15s 兜住短录音；长录音按音频长度的 0.6 倍 + -/// 10s 余量，覆盖 RTF ≤ 0.5 的机器。 -fn local_qwen_transcribe_timeout(audio_secs: f64) -> std::time::Duration { - let secs = ((audio_secs * 0.6).ceil() as u64) - .saturating_add(10) - .max(COORDINATOR_GLOBAL_TIMEOUT_SECS); - std::time::Duration::from_secs(secs) -} - -/// sherpa-onnx offline batch 暂与 Foundry 同档；后续按 Windows 真机 CPU/模型 -/// 实测结果再调整。 -#[cfg(target_os = "windows")] -fn sherpa_audio_transcribe_timeout_duration() -> std::time::Duration { - std::time::Duration::from_secs(COORDINATOR_GLOBAL_TIMEOUT_SECS) -} - /// 检查 begin_session 的 await 间隙是否被 cancel_session 打断。 /// 必须在持有 state lock 的瞬间读，结果一拿就过期，所以用 helper 名字提醒只在 /// 「准备做下一步副作用前」用。 @@ -3090,71 +2669,3 @@ fn set_phase_idle_if_session_matches(inner: &Arc, session_id: SessionId) state.phase = SessionPhase::Idle; } } - -// ─────────────────────────── audio bridge ─────────────────────────── - -struct DeferredAsrBridge { - state: Mutex, -} - -struct DeferredAsrState { - target: Option>, - pending_audio: Vec, - attaching: bool, -} - -impl DeferredAsrBridge { - fn new() -> Self { - Self { - state: Mutex::new(DeferredAsrState { - target: None, - pending_audio: Vec::new(), - attaching: false, - }), - } - } - - fn attach(&self, target: Arc) -> usize { - let mut flushed_bytes = 0; - { - let mut state = self.state.lock(); - state.attaching = true; - } - - loop { - let pending = { - let mut state = self.state.lock(); - if state.pending_audio.is_empty() { - state.target = Some(Arc::clone(&target)); - state.attaching = false; - return flushed_bytes; - } - std::mem::take(&mut state.pending_audio) - }; - flushed_bytes += pending.len(); - target.consume_pcm_chunk(&pending); - } - } -} - -impl crate::recorder::AudioConsumer for DeferredAsrBridge { - fn consume_pcm_chunk(&self, pcm: &[u8]) { - let target = { - let mut state = self.state.lock(); - if state.attaching { - state.pending_audio.extend_from_slice(pcm); - return; - } - if let Some(target) = state.target.as_ref() { - Some(Arc::clone(target)) - } else { - state.pending_audio.extend_from_slice(pcm); - None - } - }; - - if let Some(target) = target { - target.consume_pcm_chunk(pcm); - } - } -} diff --git a/openless-all/app/src-tauri/src/coordinator/asr_setup.rs b/openless-all/app/src-tauri/src/coordinator/asr_setup.rs new file mode 100644 index 00000000..fe66f248 --- /dev/null +++ b/openless-all/app/src-tauri/src/coordinator/asr_setup.rs @@ -0,0 +1,506 @@ +//! ASR 引擎构建与生命周期：麦克风/凭证前置校验、本地模型释放调度（Foundry/sherpa/ +//! local-qwen3）、QA-ASR provider 构建（QaAsrStart）、全局超时常量与各 provider 超时， +//! 以及音频线程驱动的 DeferredAsrBridge。 +//! +//! 从 `coordinator.rs` 机械拆出（行为保持，仅可见性提升）。schedule_*_release 的 +//! session-staleness 守卫与 DeferredAsrBridge 的 attach/flush 交错是并发敏感路径，只搬不改。 +//! ActiveAsr/ActiveAsrProviderKind 及其分类器留在父模块（Inner 字段类型）。 + +use super::*; + +pub(crate) fn ensure_microphone_permission(_inner: &Arc) -> Result<(), String> { + use crate::permissions::{self, PermissionStatus}; + + #[cfg(target_os = "windows")] + { + if permissions::windows_microphone_access_explicitly_denied() { + return Err("需要麦克风权限，当前状态: Denied".to_string()); + } + return Ok(()); + } + + let status = permissions::check_microphone(); + if matches!( + status, + PermissionStatus::Granted | PermissionStatus::NotApplicable + ) { + return Ok(()); + } + + // 听写路径不抢前台焦点：缺 mic 权限时直接请求系统授权，不再先 show_main_window。 + // 用户在设置页手动点“请求权限”仍走 request_microphone_from_foreground，那是显式操作。 + // 这里若系统不弹框，后续会通过 capsule error 引导用户主动去权限页处理。详见 #166。 + let requested = permissions::request_microphone(); + if matches!( + requested, + PermissionStatus::Granted | PermissionStatus::NotApplicable + ) { + Ok(()) + } else { + Err(format!("需要麦克风权限，当前状态: {requested:?}")) + } +} + +pub(crate) fn ensure_asr_credentials() -> Result<(), String> { + let active_asr = CredentialsVault::get_active_asr(); + + // 本地 Qwen3-ASR 没有"凭据"概念，但需要：(a) macOS 平台 (b) 模型已下载。 + if crate::asr::local::is_local_qwen3(&active_asr) { + #[cfg(not(target_os = "macos"))] + { + return Err("本地 ASR 当前仅支持 macOS（Windows 见 issue #256）".to_string()); + } + #[cfg(target_os = "macos")] + { + return ensure_local_qwen3_model_ready(); + } + } + + if crate::asr::local::foundry::is_foundry_local_whisper(&active_asr) { + #[cfg(not(target_os = "windows"))] + { + return Err("Foundry Local Whisper 当前仅支持 Windows".to_string()); + } + #[cfg(target_os = "windows")] + { + return Ok(()); + } + } + + if crate::asr::local::sherpa::is_sherpa_onnx_local(&active_asr) { + #[cfg(not(target_os = "windows"))] + { + return Err("sherpa-onnx local ASR 当前仅支持 Windows".to_string()); + } + #[cfg(target_os = "windows")] + { + return Ok(()); + } + } + + if is_whisper_compatible_provider(&active_asr) || is_bailian_provider(&active_asr) { + let api_key = CredentialsVault::get(CredentialAccount::AsrApiKey) + .ok() + .flatten() + .unwrap_or_default(); + if api_key.trim().is_empty() { + return Err("请先在设置中填写 ASR 服务商 API Key".to_string()); + } + return Ok(()); + } + + let creds = read_volc_credentials(); + if creds.app_id.trim().is_empty() || creds.access_token.trim().is_empty() { + Err("请先在设置中填写火山引擎 ASR App Key 和 Access Key".to_string()) + } else { + Ok(()) + } +} + +#[cfg(test)] +pub(crate) fn is_keyless_local_asr_provider(id: &str) -> bool { + if crate::asr::local::is_local_qwen3(id) { + return true; + } + #[cfg(target_os = "windows")] + { + crate::asr::local::foundry::is_foundry_local_whisper(id) + || crate::asr::local::sherpa::is_sherpa_onnx_local(id) + } + #[cfg(not(target_os = "windows"))] + { + let _ = id; + false + } +} + +#[cfg(target_os = "macos")] +pub(crate) fn ensure_local_qwen3_model_ready() -> Result<(), String> { + let prefs = || -> Result { + // 这里没法拿到 inner，直接读 preferences.json 即可（Coordinator 写盘后总是同步的）。 + crate::persistence::PreferencesStore::new() + .map_err(|e| e.to_string()) + .map(|s| s.get()) + }()?; + let model_id = crate::asr::local::ModelId::from_str(&prefs.local_asr_active_model) + .ok_or_else(|| format!("未知的本地模型 id: {}", prefs.local_asr_active_model))?; + if !crate::asr::local::models::is_downloaded(model_id) { + return Err(format!( + "本地模型 {} 未下载完整，请到 设置 → 模型设置 中下载", + model_id.as_str() + )); + } + Ok(()) +} + +/// 一次 dictation 结束后，按 prefs.local_asr_keep_loaded_secs 决定何时释放 +/// 内存里的 Qwen3-ASR 引擎。0 = 立即释放；其它值 = sleep N 秒后看 last_used。 +/// 多次会话叠加多个 sleep 任务，每个独立 check：只要中间又被使用过就跳过释放。 +pub(crate) fn schedule_local_asr_release(inner: &Arc) { + let keep_secs = inner.prefs.get().local_asr_keep_loaded_secs; + let cache = Arc::clone(&inner.local_asr_cache); + if keep_secs == 0 { + cache.release_now(); + return; + } + let dur = std::time::Duration::from_secs(keep_secs as u64); + tauri::async_runtime::spawn(async move { + tokio::time::sleep(dur).await; + cache.release_if_idle(dur); + }); +} + +#[cfg(target_os = "windows")] +pub(crate) fn foundry_local_asr_release_keep_secs(inner: &Arc) -> u32 { + inner.prefs.get().foundry_local_asr_keep_loaded_secs +} + +#[cfg(target_os = "windows")] +#[derive(Clone, Copy)] +pub(crate) enum AsrReleaseSession { + Dictation(SessionId), + Qa(SessionId), +} + +#[cfg(target_os = "windows")] +pub(crate) fn asr_release_session_is_current( + inner: &Arc, + session: AsrReleaseSession, +) -> bool { + match session { + AsrReleaseSession::Dictation(session_id) => inner.state.lock().session_id == session_id, + AsrReleaseSession::Qa(session_id) => inner.qa_state.lock().session_id == session_id, + } +} + +#[cfg(target_os = "windows")] +pub(crate) fn schedule_foundry_local_asr_release(inner: &Arc, session: AsrReleaseSession) { + let keep_secs = foundry_local_asr_release_keep_secs(inner); + let runtime = Arc::clone(&inner.foundry_local_runtime); + let inner = Arc::clone(inner); + tauri::async_runtime::spawn(async move { + if keep_secs > 0 { + tokio::time::sleep(std::time::Duration::from_secs(keep_secs as u64)).await; + } + if !asr_release_session_is_current(&inner, session) { + return; + } + if let Err(error) = runtime.release_now().await { + log::warn!("[foundry-asr] scheduled release failed: {error:#}"); + } + }); +} + +#[cfg(target_os = "windows")] +pub(crate) fn sherpa_onnx_release_keep_secs(inner: &Arc) -> u32 { + inner.prefs.get().sherpa_onnx_keep_loaded_secs +} + +/// 与 `schedule_foundry_local_asr_release` 同形：session_id 老旧则不释放， +/// 避免下一轮 session 立即重加载同一个 offline batch 模型。 +#[cfg(target_os = "windows")] +pub(crate) fn schedule_sherpa_onnx_release(inner: &Arc, session: AsrReleaseSession) { + let keep_secs = sherpa_onnx_release_keep_secs(inner); + let runtime = Arc::clone(&inner.sherpa_onnx_runtime); + let inner = Arc::clone(inner); + tauri::async_runtime::spawn(async move { + if keep_secs > 0 { + tokio::time::sleep(std::time::Duration::from_secs(keep_secs as u64)).await; + } + if !asr_release_session_is_current(&inner, session) { + return; + } + if let Err(error) = runtime.release_now().await { + log::warn!("[sherpa-asr] scheduled release failed: {error:#}"); + } + }); +} + +#[cfg(target_os = "macos")] +pub(crate) async fn build_local_qwen3( + inner: &Arc, +) -> anyhow::Result> { + let prefs = inner.prefs.get(); + let model_id = crate::asr::local::ModelId::from_str(&prefs.local_asr_active_model) + .ok_or_else(|| anyhow::anyhow!("未知本地模型 id: {}", prefs.local_asr_active_model))?; + let dir = crate::asr::local::models::model_dir(model_id)?; + let app = inner + .app + .lock() + .clone() + .ok_or_else(|| anyhow::anyhow!("AppHandle 未绑定"))?; + // 走缓存：如果已有同 id 的引擎在内存里就直接复用，避免每次会话都重加载 + // 1.2GB+ 模型。第一次加载阻塞数秒，spawn_blocking 不卡 tokio runtime。 + let cache = Arc::clone(&inner.local_asr_cache); + let mid = model_id.as_str().to_string(); + let engine = tauri::async_runtime::spawn_blocking(move || cache.get_or_load(&mid, &dir)) + .await + .map_err(|e| anyhow::anyhow!("spawn_blocking join failed: {e:#}"))??; + Ok(Arc::new(crate::asr::local::LocalQwenAsr::new(app, engine))) +} + +pub(crate) enum QaAsrStart { + Volcengine { + asr: Arc, + bridge: Arc, + }, + Bailian { + asr: Arc, + bridge: Arc, + }, + Ready { + active: ActiveAsr, + consumer: Arc, + }, +} + +impl QaAsrStart { + pub(crate) fn active_asr(&self) -> ActiveAsr { + match self { + QaAsrStart::Volcengine { asr, .. } => ActiveAsr::Volcengine(Arc::clone(asr)), + QaAsrStart::Bailian { asr, .. } => ActiveAsr::Bailian(Arc::clone(asr)), + QaAsrStart::Ready { active, .. } => active.clone(), + } + } + + pub(crate) fn recorder_consumer(&self) -> Arc { + match self { + QaAsrStart::Volcengine { bridge, .. } => Arc::clone(bridge) as _, + QaAsrStart::Bailian { bridge, .. } => Arc::clone(bridge) as _, + QaAsrStart::Ready { consumer, .. } => Arc::clone(consumer), + } + } + + pub(crate) async fn open_streaming_session(&self) -> Result<(), String> { + match self { + QaAsrStart::Volcengine { asr, bridge } => { + asr.open_session().await.map_err(|e| e.to_string())?; + let target: Arc = Arc::clone(asr) as _; + let flushed = bridge.attach(target); + log::info!("[coord] QA ASR connected; flushed {flushed} deferred audio bytes"); + Ok(()) + } + QaAsrStart::Bailian { asr, bridge } => { + asr.open_session().await.map_err(|e| e.to_string())?; + let target: Arc = Arc::clone(asr) as _; + let flushed = bridge.attach(target); + log::info!( + "[coord] QA Bailian ASR connected; flushed {flushed} deferred audio bytes" + ); + Ok(()) + } + QaAsrStart::Ready { .. } => Ok(()), + } + } +} + +pub(crate) async fn build_qa_asr_start( + inner: &Arc, + active_asr: &str, +) -> Result { + #[cfg(target_os = "windows")] + if foundry::is_foundry_local_whisper(active_asr) { + let prefs = inner.prefs.get(); + let model_alias = if foundry::model_alias_is_known(&prefs.foundry_local_asr_model) { + prefs.foundry_local_asr_model.clone() + } else { + foundry::DEFAULT_MODEL_ALIAS.to_string() + }; + let language_hint = prefs.foundry_local_asr_language_hint.trim().to_string(); + let language_hint = if language_hint.is_empty() { + None + } else { + Some(language_hint) + }; + let local = Arc::new(FoundryLocalWhisperAsr::new( + Arc::clone(&inner.foundry_local_runtime), + model_alias, + prefs.foundry_local_runtime_source.clone(), + language_hint, + )); + let active = ActiveAsr::FoundryLocalWhisper(Arc::clone(&local)); + let consumer: Arc = local; + return Ok(QaAsrStart::Ready { active, consumer }); + } + + #[cfg(target_os = "windows")] + if sherpa::is_sherpa_onnx_local(active_asr) { + let prefs = inner.prefs.get(); + let model_alias = if sherpa::model_alias_is_known(&prefs.sherpa_onnx_model) { + prefs.sherpa_onnx_model.clone() + } else { + sherpa::DEFAULT_MODEL_ALIAS.to_string() + }; + let language_hint = prefs.sherpa_onnx_language_hint.trim().to_string(); + let language_hint = if language_hint.is_empty() { + None + } else { + Some(language_hint) + }; + let token_handler = inner.app.lock().clone().map(|app| { + Arc::new(move |piece: String| { + if let Err(error) = app.emit("local-asr-token", piece) { + log::warn!("[sherpa-asr] emit token failed: {error}"); + } + }) as crate::asr::local::sherpa_provider::SherpaTokenHandler + }); + let local = SherpaOnnxAsr::new_for_model( + Arc::clone(&inner.sherpa_onnx_runtime), + model_alias, + language_hint, + token_handler, + ) + .await + .map_err(|e| format!("sherpa-onnx init failed: {e}"))?; + let local = Arc::new(local); + let active = ActiveAsr::SherpaOnnxLocal(Arc::clone(&local)); + let consumer: Arc = local; + return Ok(QaAsrStart::Ready { active, consumer }); + } + + #[cfg(target_os = "macos")] + if crate::asr::local::is_local_qwen3(active_asr) { + let local = build_local_qwen3(inner) + .await + .map_err(|e| format!("local ASR init failed: {e}"))?; + let active = ActiveAsr::Local(Arc::clone(&local)); + let consumer: Arc = local; + return Ok(QaAsrStart::Ready { active, consumer }); + } + + match active_asr_provider_kind(active_asr) { + ActiveAsrProviderKind::Bailian => Ok(QaAsrStart::Bailian { + asr: Arc::new(BailianRealtimeASR::new(read_bailian_credentials())), + bridge: Arc::new(DeferredAsrBridge::new()), + }), + ActiveAsrProviderKind::Mimo => { + let (api_key, base_url, model) = read_mimo_credentials(); + let mimo = Arc::new(MimoBatchASR::new(api_key, base_url, model)); + let active = ActiveAsr::Mimo(Arc::clone(&mimo)); + let consumer: Arc = mimo; + Ok(QaAsrStart::Ready { active, consumer }) + } + ActiveAsrProviderKind::WhisperCompatible => { + let (api_key, base_url, model) = read_whisper_credentials(); + let whisper_prompt = + crate::asr::whisper::build_prompt_from_phrases(&enabled_phrases(inner)); + let whisper = Arc::new( + WhisperBatchASR::new( + api_key, + base_url, + model, + whisper_prompt, + batch_asr_chunk_limit_ms(active_asr), + whisper_supports_verbose_json(active_asr), + ) + .with_request_format(whisper_request_format(active_asr)), + ); + let active = ActiveAsr::Whisper(Arc::clone(&whisper)); + let consumer: Arc = whisper; + Ok(QaAsrStart::Ready { active, consumer }) + } + ActiveAsrProviderKind::Volcengine => Ok(QaAsrStart::Volcengine { + asr: Arc::new(VolcengineStreamingASR::new( + read_volc_credentials(), + enabled_hotwords(inner), + )), + bridge: Arc::new(DeferredAsrBridge::new()), + }), + } +} + +/// Coordinator 全局超时保护：防止 ASR await_final_result() 永远挂起。 +/// 设置为 15 秒（比 ASR 的 12 秒 FINAL_RESULT_TIMEOUT 稍长）， +/// 只在 ASR 超时机制失效时作为最后的防线触发。 +pub(crate) const COORDINATOR_GLOBAL_TIMEOUT_SECS: u64 = 15; + +#[cfg(target_os = "windows")] +pub(crate) fn foundry_audio_transcribe_timeout_duration() -> std::time::Duration { + std::time::Duration::from_secs(COORDINATOR_GLOBAL_TIMEOUT_SECS) +} + +/// 本地 Qwen3-ASR 的动态转写超时。固定 15 秒在长录音（≥ 30s）+ 慢机器 +/// （RTF ≈ 0.3–0.5）上必然超时把整段内容丢掉。改用 max(15, ceil(audio_s +/// × 0.6) + 10)：基础保留 15s 兜住短录音；长录音按音频长度的 0.6 倍 + +/// 10s 余量，覆盖 RTF ≤ 0.5 的机器。 +pub(crate) fn local_qwen_transcribe_timeout(audio_secs: f64) -> std::time::Duration { + let secs = ((audio_secs * 0.6).ceil() as u64) + .saturating_add(10) + .max(COORDINATOR_GLOBAL_TIMEOUT_SECS); + std::time::Duration::from_secs(secs) +} + +/// sherpa-onnx offline batch 暂与 Foundry 同档；后续按 Windows 真机 CPU/模型 +/// 实测结果再调整。 +#[cfg(target_os = "windows")] +pub(crate) fn sherpa_audio_transcribe_timeout_duration() -> std::time::Duration { + std::time::Duration::from_secs(COORDINATOR_GLOBAL_TIMEOUT_SECS) +} + +// ─────────────────────────── audio bridge ─────────────────────────── + +pub(crate) struct DeferredAsrBridge { + state: Mutex, +} + +pub(crate) struct DeferredAsrState { + target: Option>, + pending_audio: Vec, + attaching: bool, +} + +impl DeferredAsrBridge { + pub(crate) fn new() -> Self { + Self { + state: Mutex::new(DeferredAsrState { + target: None, + pending_audio: Vec::new(), + attaching: false, + }), + } + } + + pub(crate) fn attach(&self, target: Arc) -> usize { + let mut flushed_bytes = 0; + { + let mut state = self.state.lock(); + state.attaching = true; + } + + loop { + let pending = { + let mut state = self.state.lock(); + if state.pending_audio.is_empty() { + state.target = Some(Arc::clone(&target)); + state.attaching = false; + return flushed_bytes; + } + std::mem::take(&mut state.pending_audio) + }; + flushed_bytes += pending.len(); + target.consume_pcm_chunk(&pending); + } + } +} + +impl crate::recorder::AudioConsumer for DeferredAsrBridge { + fn consume_pcm_chunk(&self, pcm: &[u8]) { + let target = { + let mut state = self.state.lock(); + if state.attaching { + state.pending_audio.extend_from_slice(pcm); + return; + } + if let Some(target) = state.target.as_ref() { + Some(Arc::clone(target)) + } else { + state.pending_audio.extend_from_slice(pcm); + None + } + }; + + if let Some(target) = target { + target.consume_pcm_chunk(pcm); + } + } +}