(HJj7 2ddlZddlZddlZddlZddlZddlZddlZddlmZddlm Z m Z m Z ddl m ZddlmZddlmZddlmZdde e efdZded efd Zded efd Zd Zd efdZd efdZd efdZd efdZej>ej@Z!ejDddejFde e ejHfdZ% dde e eejFfde eejLfd edede'def dZ(dZ)e*dk(re)yy)N)pprint)ListOptionalUnion) tree_reduce) load_modelargvcztjd}|jddtd|jdtdd |jd tdd |jd tdgdd|jddd|jdtdd|jddtddd|jdtd d!|jd"t d#d$|jd%td&d'|jd(dd)|jd*tdd+|jd,td-d.|jd/tdd0|jd1t jdd2|jd3td4d5|j|S)6Nz(Generate transcriptions from audio files) descriptionz--modelz$mlx-community/whisper-large-v3-turbozPath to the model)defaulttypehelpz--audioTzPath to the audio file)r requiredrz --output-pathzPath to save the outputz--formattxt)rsrtvttjsonz&Output format (txt, srt, vtt, or json))r r choicesrz --verbose store_truezVerbose output)actionrz --max-tokensi z(Maximum number of new tokens to generate)r r rz--max-parallel-segments batch_sizeSEGMENTSzcMaximum number of audio segments to transcribe in parallel for models that support segment batching)destr r metavarrz --languageenz)Language code (e.g. en, es, fr, de, etc.)z--chunk-durationg>@z)Chunk duration in seconds (default: 30.0)z--frame-thresholdzFrame threshold (default: 25)z--streamzF 6    d1I cD7P  - 5   L?OP   7   !   3    8     8     ,    K    N     0     O    ZZ X     :    T ""secondsreturnct|dz}t|dzdz}|dz}|dd|dd|djddS)z2Convert seconds to HH:MM:SS,mmm format for SRT/VTTi<02d:z06.3f.,)r"replace)r)hoursminutess r'format_timestampr4ssV 4 E7T>b()GlGC['#a 7 ? ?S IIr(c8t|jddS)z.Convert seconds to HH:MM:SS.mmm format for VTTr0r/)r4r1)r)s r'format_vtt_timestampr6{s G $ , ,S# 66r(cht|dr<|jDcgc]&}|j|j|jd(c}Sg}|j D]S}|j |d|d|ddd|vs&|ds,|dD] }|j |d|d|dd"U|Scc}w)zExtract unified cues from any model's output (Parakeet or Whisper). Uses word-level cues when available, otherwise falls back to segment-level. sentences)startendtextr9r:r;wordsword)hasattrr8r9r:r;segmentsappend)r?scuesws r' _get_cuesrDs x%FNFXFX FXaggaeeQVV r:r; ) rKrLrMrNrO enumeraterDrPr4r?rErQicues r' save_as_srtr\s #  }D !39  # #CJJ / 0  ( 3Q7FAs GGqcH  GG#CL12%8HU8T7UUWX  GGs6{m4( ) 8  0 0 0s A8B<<Cc |dk7rt|dddn"tjtj5}|j dt t|dD]_\}}|j |d|j t|d d t|d d|j |d d a dddy#1swYyxYw)NrGz.vttrCrHrIzWEBVTT rTrUr9rVr:r;rW) rKrLrMrNrOrPrXrDr6rYs r' save_as_vttr^s #  }D !39  # #CJJ / 0    ( 3Q7FAs GGqcH  GG'G 56er;r8r9r:r`rarXrbr?r@rKrLrMrNrOrdump)r?rErAtresultrZsegrQs r' save_as_jsonrjsx%MM""++! ,AFFWW55 ! "#"*A %&FF%&WW#$55()  "* ,! ,h001DAqq,'78||{#A&|42 MM ""A& 7xeHqz1 C!|'  zG q $%lOL! :  % %c *#" #  }E "C':  # #CJJ / 0  &!%:  0 0M X 0 0s#?F4&6F/F4 F:/F4:Gmodelstreamsc#bKtjjs dytd|d}tjj d}|d|zkDr|dz}|dz}t d|d|d tj |} d||D]}tj|ntjtj |y#wxYw#||D]}tj|ntjtj |wxYww) a A context manager to temporarily change the wired limit. Note, the wired limit should not be changed during an async eval. If an async eval could be running pass in the streams to synchronize with prior to exiting the context manager. NcXt|tjr||jzS|SN) isinstancemxarraynbytes)accxs r'zwired_limit..s!Z288-D3>M#Mr(r max_recommended_working_set_sizeg?iz0[WARNING] Generating with a model that requires z6 MB which is close to the maximum recommended size of z MB. This can be slow. See the documentation for possible work-arounds: https://github.com/ml-explore/mlx-lm/tree/main#large-models)rqmetal is_availabler device_infoprintset_wired_limit synchronize)rkrl model_bytes max_rec_sizemodel_mb max_rec_mb old_limitrAs r' wired_limitrs' 88 "   ! MuVW xx++-.PQ |+ +"e+H%.J B8*MEEOLQNN  &&|4  * " ANN1%!   y )1 &" ANN1%!   y )s6D/CA)D/C"A D/CD/"A D,,D/audioformatverboser;c ddlm}| tdt|tr t |}t jtj}|r5tdtd|td|td||jd d} | r|j| |r||d <tj|j} |jD cic]\} } | | j vs| | }} } |j#d d rg} d }d}d}d}|j|fd|i|D]}|j$|j&|j(|j*d}| j-|||j$z }|j.}t1|dr|j2dkDr |j2}t1|ds|j4dkDs|j4}tj}||z }||j7| |||||z||dkDr||z nd|dkDr||z nd }n|j|f|t8d|}|rqt1|d r&tdt|j$dddt1|dr3|j:'tdt=|j:dddgztj}|rtdtd|d|td ||z d!d"t||rNtd#|j2d$|j>d%d&td'|j4d$|j@d%d&td(t jBd)z d!d*tEjFtDjHjKtDjHjM|d+,t1|dxr|j:du}t1|d-xr|jNdu}|d.k(s|s|s|s |s td/tQ|||S|d0k(rtS|||S|d1k(rtU|||S|d2k(r tW|||Scc} } w)3a Generate transcriptions from audio files. Args: model: Path to the model or the model instance. audio: Path to the audio file (str), or audio waveform (mx.array). output_path: Path to save the output. format: Output format (txt, srt, vtt, or json). verbose: Verbose output. text: Text to align (for forced alignment models). **kwargs: Additional arguments for the model's generate method. Returns: segments: The generated transcription segments. rT) STTOutputNz.Model path or model instance must be provided.z ==========zAudio path: zOutput path: zFormat: gen_kwargsr;streamFrrrr)r;r9r:is_final prompt_tokensgeneration_tokens) r;r?languagerr total_tokens total_time prompt_tpsgeneration_tps)rgeneration_streamzTranscription: iz... r?zSegments: z...z ==========zSaving file to: ./r/zProcessing time: z.2fz secondszPrompt: z tokens, z.3fz tokens-per-seczGeneration: zPeak memory: geAz GBT)exist_okr8rz2[WARNING] No segments found, saving as plain text.rrr), models.baser ValueErrorrpr!rrqreset_peak_memorytimer{popupdateinspect signaturegenerateitems parametersgetr; start_timeend_timerr@rr>rrstriprr?rrrget_peak_memoryosmakedirspathdirnameabspathr8rRr\r^rj)rkrrErrr;kwargsrrrrkv all_segmentsaccumulated_textrrrrh segment_dictstream_end_timestream_durationr?r has_segments has_sentencess r'generate_transcriptionrs.' }IJJ%5!J h +E734 ,[M:; 'x01L$/J j! v!!%..1I%||~ K~tq!i6J6J1Jad~F K zz(E"  $ennUFGFvFF **"OO L    -  + Hv/F4H4H14L & 4 4 v238P8PST8T$*$<$<!#G&))+)J6!'')!'/&)::&:IA:M}6ST7F7J!O3PQ  "5>>  "6G KQ  8V $ 3 4 X]]4C()/ 0 8Z (X->->-J . / 8$$Ra(E72 3yy{H o 1+axHI 0J1Fs0K8TU h * )(*@*@)A&&s+?<  -h.H.H-I**3/@  ,R-?-?-AC-G,LCPQKK <=M8Z0RX5F5Fd5RLHk2Ux7I7IQU7UM |MM F GHk* O 5Hk* O 5Hk* O 6 X{+ OwLs ,Q6Q6c@t}tdit|y)N)r%rvars)argss r'mainrs rs ((!*^#Xd3i(^#BJeJJ7%7C7 $s *s * *s *4;4;p"BMM"3""3"3"56 $*ryy$*8DO+D$*$*P.2"&# M E#ryy.) *M bhh MM  M  M  M`)  zFr(