iY8dZddlZddlZddlZddlmZddlmZddlZddl m Z ddl Z ddl Z ddl Z ddlmZmZmZmZmZmZmZmZddlmZddlmZdd lmZd d lmZd d lm Z m!Z!d d l"m#Z#d dl$m%Z%m&Z&d dl'm(Z(d dl)m*Z*m+Z+m,Z,m-Z-m.Z.m/Z/d dl0m1Z1m2Z2m3Z3m4Z4m5Z5m6Z6dZ7ejpe9ee7ejtejve<Z=e=j}e9ee7ejte e?jjd dz ZB eBjdd5ZDejeDZFdddeHeFjddZJeHeFjdeJZKeLeFjddZMeNeFjddjjdvZQd eRfd!ZSd"ZTed#efd$ZUeT\ZVZVZVZWed%d&eU'ZXd e#fd(ZYd)ed*eeNd+eeNd eeNfd,ZZed-fd.ed efd/Z[eXjd0ed-ededfd)ed.ed1eeNd2eeNd ef d3Z]eXjd4ed-ededfd)ed.ed1eeNd2eeNd ef d5Z^eXjd6d7eNd efd8Z_eXjd9d:efd;ZaeXjd<ed-d=d d>d?fd.ed@eHdAeHdBeHdCeNd ef dDZby#1swYxYw#eG$riZFYwxYw)EuNgzzm 服务主入口。 职责概览： - 暴露 HTTP `/transcribe` 与 WebSocket `/ws/transcribe` 两个转写接口。 - 在请求入口做参数校验、会话级资源初始化（SpeakerRegistry、队列）和结果汇总。 - 调用 services 中的统一流水线完成 ASR、二次切分、说话人匹配与时间戳合并。 N)asynccontextmanager)Optional)Path)FastAPIFileForm HTTPExceptionRequest UploadFile WebSocketWebSocketDisconnect) JSONResponse)WebSocketState) AudioSegment) choose_device) get_model init_model)SpeakerRegistry)run_async_transcribe_task run_transcribe_pipeline_for_file)transcribe_audio_file) QueueChunkSessionChunkQueuemerge_adjacent_same_speakerpreload_persisted_speakersprocess_chunk_pipelinefilter_time_stamps)create_task_recordget_task_record init_task_dbmark_task_failedmark_task_runningmark_task_succeededINFO)levelzgzzm_config.jsonrutf-8encoding chunk_seconds<ws_chunk_secondsws_emit_secondsg@enable_detailed_logF)1trueyesonreturnc tjdd5}tj|cdddS#1swYyxYw#t$ricYSwxYw)Nr'r(r)) _config_pathopenjsonload Exception)_fs /app/gzzm/app.py_load_gzzm_configr=HsF   sW  599R=6 5 5  s*A8 AAAA AAczt}t|jdd}t|jdd}|jd|jdd}t|}t |jdxsd j }t |}|d k(rd }n |d k(rd }nd}||||fS)zReturn (similarity_threshold, soft_margin, disable_soft_reuse, speaker_device). Values are read from environment variables first, then from `gzzm_config.json`. This is intentionally lightweight so callers can call it per-request. speaker_sim_thresholdg?speaker_sim_soft_margingffffff?speaker_disable_soft_reuse disable_soft_reuse_for_persistedFmodel_device_speakernpuznpu:0cudazcuda:0cpu)r=floatgetboolstrstripr)cfgsim_thrsim_soft disable_softcfg_speaker_deviceprimaryspeaker_devices r<_get_runtime_speaker_settingsrTPs  CCGG3S9:GSWW6=>H777Acej9klL %LSWW%;<BCIIK./G%  F ! HlN ::_c>KttdywN)rr!)rVs r<lifespanrYisLN sz Qwen3 ASR APIz1.0.0)titleversionrYc>t\}}}}t||||S)N)similarity_threshold soft_margin model_devicerB)rTr)rNrOrPrSs r<_build_speaker_registryr`xs,6S6U3GX|^ $#)5  rUrequestspeaker_ids_camelspeaker_ids_snakecv|xs|}|r|S|j}|jdxs|jdS)N speakerIds speaker_ids) query_paramsrI)rarbrcmanual_speaker_idsqps r<_resolve_manual_speaker_idsrjs?*?.?!!   B 66,  8266-#88rU.filecKt}| tdd|jxsdj}|j ds|j ds tdd |j drdnd}t j d | 5}|j|jd{|j}ddd tt}t| tj|S7N#1swY=xYw#t$rYSwxYw# tjw#t$rYwwxYwxYww) z7Legacy helper: pure ASR without speaker identification.NModel is not initialized status_codedetailrD.wav.mp3#Only wav or mp3 files are supportedFdeletesuffix)rr filenamelowerendswithtempfileNamedTemporaryFilewritereadnamerTRANSCRIBE_CHUNK_SECONDSrosremover:)rkmodelryrxtmptmp_pathresults r<transcribe_audio_file_handlerrs KE }4NOO #**,H   f %):):6)B4YZZ((0VfF  $ $E& AS  #$88 B&x1IJF#  IIh $ B A     IIh    sBE D4D 5D E D -DE DD E  DE DE  E"D87E8 EEEEE  /transcribererfc Kt}| tdd|jxsdj}|j ds|j ds tdd |j drdnd}t j d | 5}|j|jd{|j}dddt|||} tjd | || t| ttt t"td } t%|  t'j(|S7#1swYyxYw#t*$rYSwxYw#t$rt*$r0} tj-dtdt/| d} ~ wwxYw# t'j(w#t*$rYwwxYwxYww)zgQueue-based /transcribe pipeline: preload speakers -> chunk queue -> ASR -> second split -> speaker id.NrmrnrorDrrrsrtruFrvzT/transcribe manual speakerIds resolved: %s (form speakerIds=%s, form speaker_ids=%s)r)rrhr+rSbuild_speaker_registrydetailed_log_enabledloggerlog_tagzTranscribe failed)rr ryrzr{r|r}r~rrrjrinforrSPEAKER_DEVICEr`DETAILED_LOG_ENABLEDrrrr: exceptionrK) rarkrerfrryrxrrrhpayloades r<transcribe_audiors KE }4NOO #**,H   f %):):6)B4YZZ((0VfF  $ $E& AS  #$88 B5Wj+V KK^ 212)#:!5!  G$  IIh ?$ B AB     <,-CF;;<  IIh    sBG E4D? 5E -G 71E)E >G ?EE G  EG EG F.+FFF!!G#F98G9 GGGGG z/transcribe/asynccKt}| tdd|jxsdj}|j ds|j ds tdd |j drdnd}t j d | 5}|j|jd{|j}dddt|||} tjd | ||ttj} t!| |jxsd| t+j,t/| | t0t2t4t6t8t:t<t t?| dd| dS7#1swYxYw#t"$rP} t%j&n#t"$rYnwxYwtj)d tdd| d} ~ wwxYww)z>Submit async transcribe task and return task UUID immediately.NrmrnrorDrrrsrtruFrvzZ/transcribe/async manual speakerIds resolved: %s (form speakerIds=%s, form speaker_ids=%s)z-Failed to create async transcribe task recordrzfailed to create task: ) task_idrrh mark_runningmark_succeeded mark_failedr+rSrrrqueuedz/transcribe/async/)rstatus query_path) rr ryrzr{r|r}r~rrrjrrrKuuiduuid4rr:rrrasyncio create_taskrr#r$r"rrr`rr) rarkrerfrryrxrrrhrrs r<transcribe_audio_asyncrs KE }4NOO #**,H   f %):):6)B4YZZ((0VfF  $ $E& AS  #$88 B5Wj+V KKd $**,GS7DMM$7R9KL !1*.(2)#:!5  .wi8  M$ B A S  IIh    HI6MaS4QRR SstBG4F 4F 5F  A G4F/AG4 F  FG4 G1"F87G,8 GG,G(G,,G11G4z/transcribe/async/{task_id}rcTKt|}| tddt|Sw)z6Query async transcribe task status and result by UUID.iztask not foundro)r r r)rtasks r<get_transcribe_taskrs0 7 #D |4DEE  s&(z/ws/transcribe websocketc Kjd{t}|1tjdj ddd{y j d{}|jd d xsd }t|jd d td t|jdd t|jdd|d k7r2tjd|j ddd{ytd zz}tdtt}td tt|t|z}tt}td|vr|jdn|jd}tj!d|t#|d} ddgddt dt$fddtdt$ddffd } dt&dt(ffd dt*t(t,tt,t$fff d! } j/d{} | jd"d#k(r8tj!d$| jd| jd| n| jd"d%k(r| jd&| jd&} |j1| t3||k\rbt5|d|}|d|=j7t9| |tt3|t|z '| d z } t3||k\rb| d{\}}}|s||r| ||d{n| jd"d%k(r| jd(r| jd(d)k(r|r[t5|}|j;j7t9| |tt3|t|z '| d z } | d{\}}}|s+tjd*||r| ||d{n*d+dkDrd z ndtd,d-d{H j;y77Y7B#t $r4tjdj ddd{7YywxYw7777k777q#t<$r6}tj!d.|j>tA|ddYd}~d}~wt $r[tjCd/ tjd0j d1d2d{7n#t $rYnwxYwY wxYw#t $rYywxYw# j;w#t $rYwwxYwxYww)3zqQueue-based WS pipeline. Client controls session lifecycle; server keeps connection open unless error/disconnect.Nz5WS closed before processing: model is not initializedirncodereasonz7WS closed due to invalid init payload: JSON is requirediz&Invalid init payload, JSON is requiredformat pcm_s16le sample_rate>rchannels sample_widthz'WS closed due to unsupported format: %szOnly pcm_s16le is supportedg?rerfz$/ws/transcribe manual speakerIds: %srgrr4cRt|j}|r|St|SrX)rKrLrepr)rtexts r< _format_excz)websocket_transcribe.._format_excWs"1v||~ KAwrUrrc hKtddtjk(rtj d||y j ||d{tj d||y7#t$r6}tj d||t|j|Yd}~yd}~wwxYww)Napplication_statezJWS close skipped because application_state=DISCONNECTED: code=%s reason=%srz&WS closed by server: code=%s reason=%szFWS close skipped/failed: code=%s reason=%s error_type=%s error_repr=%r) getattrr DISCONNECTEDrrclosewarningr:type__name__)rr close_errorrs r< safe_closez(websocket_transcribe..safe_close]s 914 8N.safe_send_jsonms %%g. . . /"  KK~ F# G$8T* >48 #6=   NNe F# G$Q  A >48 #6=  sJD!!D! DA$BD DA(D D DDc  Kjsj}| ytr0tj d|j |j ztjtj|j }tj} |jdt|t t"d{}|j'd |j'd xsd j)}|j'd xsg} t+|}t-|}|r j/|t-  t1|j'd|j xs |j } tr'tj d|j | z|d||t3| dt3| zddd{} | s't5ddt6j8k(rydddfS| z dz jsy7Y#t$$rA} |}d|dd{7}|s dddfcYd}~Sdddd |fcYd}~Sd}~wwxYw#t$$rYYwxYw7w)NzI[detailed][/ws/transcribe] start chunk index=%s seq=%s range=[%.3f, %.3f])r frame_rater chunk_error)rrerrorFz!chunk_error send failed at index=z!chunk processing failed at index=z: languagerrD time_stamps duration_seczT[detailed][/ws/transcribe] result chunk index=%s seq=%s range=[%.3f, %.3f] result=%s chunk_result)rrrrchunk_durationtotal_durationr)FNz.client disconnected while sending chunk_resultz"chunk_result send failed at index=r)TNN)emptytry_getrrrseqrrfrom_rawioBytesIOrrget_running_looprun_in_executorrWS_CHUNK_SECONDSrr:rIrLrrextendrHroundrrr)itemsegmentlooprr error_textsent chunk_textchunk_tsrokrall_time_stampsrrr offset_secqueuerrrspeaker_registryrs r< process_queuez+websocket_transcribe..process_queues++-==?D|f c$ _HH!2!22 #++ 4<<()&! G++-D ^%)%9%9*$$"&  '++J7&**628b??AJ#'' 6<"H -h738EkG.06Bd6JDHH\2PTPXPXYfPg KK68MN C EJOHyS sCD dt6Y tXc]HSM'I!JY Y Y vM!))++Cwwv"88 SGGFOGGH%  wwv"55#''':J:Vwww't$'lk1#GL[$9:G  -II" #$+).s7|)V??W'W$WW$ W!W$ W!!W$$W'z/speaker/registerrrrrrrrcK|jd{}d}|jxsdj}|dvs|jdra t j t j|}t j} |j| d| j}n7|d vs|jd r|d k7r tdd |}n tdd t} | j|||t} t#d| iS7#t$r} tdd|  d} ~ wwxYw#t$rtdd t$r} tdt!|  d} ~ wwxYww)z.Persist one speaker embedding and return UUID.NrD)wavmp3)rrrsraw)rrtzfailed to decode audio: ro)pcmrz.pcmrz-only 16-bit PCM (sample_width=2) is supportedz6unsupported audio format; provide wav/mp3 or pcm_s16le)rrdeviceiz2speechbrain not available for embedding extractionrr)rryrzr{r from_filerrexportgetvaluer:r r`persist_from_pcmr ImportErrorrKr) rkrrrrr pcm_bytesfnamesegbufrreguids r<speaker_registerr3sk DI ]] b ' ' )E 5>>2B#C X((D)9:C**,C JJs5J ) I ) )U^^F-Cq #r1s *  kkk*/-(-Y-  ''9gllCD   8 $GLL9:H~%%'//25GG    3  1R tyy}  2|//DE|''(: