uPjgFUdZddlZddlZddlmZmZddlmZddlm Z m Z m Z m Z ddl mZmZddlmZddlmZmZmZdd lmZmZdd lmZmZdd lmZe rdd lmZ d Z!e e"d<e!dzZ#e e"d<dZ$Gdde Z%Gdde Z&Gdde Z'Gdde Z(Gdde Z)Gdde Z*Gdde Z+d ed!ed"e%ffd#Z,d$ed!e+fd%Z-Gd&d'Z.d!e/eeffd(Z0y))a] NVIDIA Riva STT handler. This module bridges litellm's transcription dispatch to NVIDIA Riva's gRPC streaming ASR API. We do *not* go through ``base_llm_http_handler`` because Riva is gRPC-only: HTTP-shaped abstractions (``httpx.Response``, ``api_base/v1/...`` URLs, multipart bodies) do not apply. The handler is intentionally a thin orchestration layer: 1. Resample the inbound audio to 16 kHz mono LINEAR_PCM (Riva's required wire format). 2. Build ``RecognitionConfig`` / ``StreamingRecognitionConfig`` protobufs from the structured dict produced by :class:`NvidiaRivaAudioTranscriptionConfig`. 3. Construct ``riva.client.Auth`` honoring NVCF (function-id metadata + TLS) vs self-hosted (any host:port, optional TLS) modes. 4. Stream the audio through Riva's ``streaming_response_generator`` and aggregate ``is_final`` results into a single transcript. 5. Return a normalized ``TranscriptionResponse`` with ``duration`` exposed on ``_hidden_params`` so cost calculation works. ``riva-client`` is imported lazily so ``litellm`` core remains usable without the optional STT extras installed. N)CallableIterable) ModuleType) TYPE_CHECKINGAnyFinalProtocol)get_audio_file_nameprocess_audio_file)resample_to_riva_pcm)RIVA_TARGET_NUM_CHANNELSRIVA_TARGET_SAMPLE_RATE_HZ"NvidiaRivaAudioTranscriptionConfig)NvidiaRivaExceptiongrpc_error_to_litellm_exception) FileTypesTranscriptionResponse) convert_to_model_response_object)Loggingi@_DEFAULT_CHUNK_SAMPLES_DEFAULT_CHUNK_BYTESz[NVIDIA Riva client is not installed. Install with `pip install 'litellm[stt-nvidia-riva]'`.ceZdZdZy) _RivaAuthz#Opaque ``riva.client.Auth`` handle.N__name__ __module__ __qualname____doc__y/home/oem/CyberBrain26-unified/.venv/lib/python3.12/site-packages/litellm/llms/nvidia_riva/audio_transcription/handler.pyrrCs-r!rc2eZdZededeeffdZy) _AsrServicereturn.cyNr selfs r"streaming_response_generatorz(_AsrService.streaming_response_generatorHsORr!N)rrrpropertyrrobjectr*r r!r"r$r$Gs$ RhsHV i(DKDDU}UF#'#>#>v#FM B^^'+&77$;  '+=&9%j124KL2  #0,5,F,F 8X0"0'/ 8 E#   <1!4! ; s"#2"5"56H"I"1"5"5i"@3C3O./UYZjUk13  OO]C0@,AB C  OO_y.AB C Q-=-(G[cd d Q-=-dGXxP P Qs4BB('B(rArgc|jdxsdj}tt||t|j}|j |t |d|dt |dt|jddt|jdd t |jd d |jd d xsd t|jdd t|jdd  }|jd}t|tr2|r0 |jdi|}|jj||S|S#t$rY|SwxYw)Nrcr<ra language_coderbenable_automatic_punctuationTenable_word_time_offsetsFmax_alternativesrGverbatim_transcriptsprofanity_filter) rcrarrbrrrrGrrr6r )ruppergetattrrDr<RecognitionConfigintrrrEndpointingConfigr6r1r)r)rArg encoding_name encoding_enumrh endpointingeps r"rzPRW"XY-N - 5889MN k4 ([ 0Q0Q0Q0`T_0`))2226 v    s8-E)) E65E6 callable_obj.c tj|}|j}d|vryt d|j DS#ttf$rYywxYw)NFrLTc3jK|]+}|jtjjk(-ywr')kindinspect Parameter VAR_KEYWORD).0ps r" zGNvidiaRivaAudioTranscription._supports_timeout_kwarg..s%Tq166W..:::Ts13)r signaturer ValueError parametersanyvalues)rsigparamss r"rz4NvidiaRivaAudioTranscription._supports_timeout_kwargs]  **<8C  TFMMOTTT :&  sAAArc#pKtdt|tD]}|||tz}|s|yw)Nr)rangelenr)roffsetchunks r"rz/NvidiaRivaAudioTranscription._iter_audio_chunkss@As9~/CD Ffv0D'DEEK  s46cg}|D]}t|ddxsg}|D]}t|ddst|ddxsg}|s%|d}t|ddxsd}t|d dxsg}g} |D]R} | jt| d dtt| d dxsdtt| d dxsdd T|j|| d|S)a Walk the gRPC stream, ignore empty / non-final chunks, and return a list of normalized final-result dicts. Matching the user's note: the ``id`` blocks with no ``results`` are streaming heartbeats and must be skipped. resultsNis_finalF alternativesr transcriptrwordsword start_timeend_time)r start_time_ms end_time_ms)rr)rrr) rrrresponserresultrtopr words_protorrs r"rz3NvidiaRivaAudioTranscription._collect_final_resultss9;  QHh 48>BG! Qvz59&v~tDJ #"1o$S,;Ar %c7D9?R 'DLL$+D&"$=-0|Q1O1TST-U+.wtZ/K/Pq+Q$$J%OP% Q Q*r!)FNr')F)rrrrrrrrrrrrWrUrVrrrrr4r staticmethodrr,rbytesrlistrr r!r"rFrFjs0 %EI+ + +  +  + . + + )+ t+ *+ + rs4.66 A:!%$4q8e8s..S(S 993h3AA B(B(X( zhsI~.F)Z)N)JJZ (eJ 23(r!