openvidu-local-deployment/community/agent-speech-processing.yaml

##############################
# Agent common configuration #
##############################

# Docker image of the agent.
docker_image: docker.io/openvidu/agent-speech-processing:main

# Whether this agent should run or not, and how it will connect to Rooms (automatic|manual|disabled)
# automatic: the agent will run and will automatically connect to new Rooms.
# manual: the agent will run and will connect to new Rooms only when your application dictates it by using the Agent Dispatch API.
# disabled: the agent will not run.
processing: manual

################################
# Agent specific configuration #
################################
speech_to_text:
  # Whether or not the agent should be hidden to the Participants of the Room.
  hidden: true

  # Which Speech-To-Text provider to use: aws, azure, google, opeanai, groq, deepgram, assemblyai, fal, clova, speechmatics
  # The custom configuration for the selected provider must be set below
  provider: aws

  aws:
    # Credentials for AWS Transcribe. See https://docs.aws.amazon.com/transcribe/latest/dg/what-is.html
    aws_access_key_id:
    aws_secret_access_key:
    aws_default_region:
    # See https://docs.aws.amazon.com/transcribe/latest/dg/supported-languages.html
    language:
    # The name of the custom vocabulary you want to use.
    # See https://docs.aws.amazon.com/transcribe/latest/dg/custom-vocabulary.html
    vocabulary_name:
    # The name of the custom language model you want to use.
    # See https://docs.aws.amazon.com/transcribe/latest/dg/custom-language-models-using.html
    language_model_name:
    # Whether or not to enable partial result stabilization. Partial result stabilization can reduce latency in your output, but may impact accuracy.
    # See https://docs.aws.amazon.com/transcribe/latest/dg/streaming-partial-results.html#streaming-partial-result-stabilization
    enable_partial_results_stabilization:
    # Specify the level of stability to use when you enable partial results stabilization (enable_partial_results_stabilization: true). Valid values: high | medium | low
    # See https://docs.aws.amazon.com/transcribe/latest/dg/streaming-partial-results.html#streaming-partial-result-stabilization
    partial_results_stability:
    # The name of the custom vocabulary filter you want to use to mask or remove words.
    # See https://docs.aws.amazon.com/transcribe/latest/dg/vocabulary-filtering.html
    vocab_filter_name:
    # The method used to filter the vocabulary. Valid values: mask | remove | tag
    # See https://docs.aws.amazon.com/transcribe/latest/dg/vocabulary-filtering.html
    vocab_filter_method:

  azure:
    # Credentials for Azure Speech Service.
    # One of these combinations must be set:
    #  - speech_host
    #  - speech_key + speech_region
    #  - speech_auth_token + speech_region
    # See https://learn.microsoft.com/en-us/azure/ai-services/speech-service/get-started-speech-to-text?tabs=macos%2Cterminal&pivots=programming-language-python#prerequisites
    speech_host:
    speech_key:
    speech_auth_token:
    speech_region:
    # Azure handles multiple languages and can auto-detect the language used. It requires the candidate set to be set.
    # See https://learn.microsoft.com/en-us/azure/ai-services/speech-service/language-support?tabs=stt#supported-languages
    languages:
    # Removes profanity (swearing), or replaces letters of profane words with stars. Valid values: Masked | Removed | Raw
    # See https://learn.microsoft.com/en-us/azure/ai-services/translator/profanity-filtering
    profanity:

  google:
    # Credentials for Google Cloud. This is the content of a Google Cloud credential JSON file.
    # Below is a dummy example for a credential type of "Service Account" (https://cloud.google.com/iam/docs/service-account-creds#key-types)
    credentials_info: |
      {
        "type": "service_account",
        "project_id": "my-project",
        "private_key_id": "xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",
        "private_key": "-----BEGIN PRIVATE KEY-----\nxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx\n-----END PRIVATE KEY-----\n",
        "client_email": "my-email@my-project.iam.gserviceaccount.com",
        "client_id": "xxxxxxxxxxxxxxxxxxxxx",
        "auth_uri": "https://accounts.google.com/o/oauth2/auth",
        "token_uri": "https://oauth2.googleapis.com/token",
        "auth_provider_x509_cert_url": "https://www.googleapis.com/oauth2/v1/certs",
        "client_x509_cert_url": "https://www.googleapis.com/robot/v1/metadata/x509/my-email%40my-project.iam.gserviceaccount.com",
        "universe_domain": "googleapis.com"
      }
    # Which model to use for recognition. If not set, uses the default model for the selected language.
    # See https://cloud.google.com/speech-to-text/docs/transcription-model
    model:
    # The location to use for recognition. Default is "us-central1". Latency will be best if the location is close to your users.
    # Check supported languages and locations at https://cloud.google.com/speech-to-text/v2/docs/speech-to-text-supported-languages
    location:
    # List of language codes to recognize. Default is ["en-US"].
    # See https://cloud.google.com/speech-to-text/v2/docs/speech-to-text-supported-languages
    languages:
    # Whether to detect the language of the audio. Default is true.
    detect_language:
    # If 'true', adds punctuation to recognition result hypotheses. This feature is only available in select languages. Setting this
    # for requests in other languages has no effect at all. The default 'false' value does not add punctuation to result hypotheses.
    # See https://cloud.google.com/speech-to-text/docs/automatic-punctuation
    punctuate:
    # The spoken punctuation behavior for the call. If not set, uses default behavior based on model of choice.
    # e.g. command_and_search will enable spoken punctuation by default. If 'true', replaces spoken punctuation
    # with the corresponding symbols in the request. For example, "how are you question mark" becomes "how are you?".
    # See https://cloud.google.com/speech-to-text/docs/spoken-punctuation for support. If 'false', spoken punctuation is not replaced.
    spoken_punctuation:
    # Whether to return interim (non-final) transcription results. Defaults to true.
    interim_results:

  openai:
    # API key for OpenAI. See https://platform.openai.com/api-keys
    api_key:
    # See https://platform.openai.com/docs/guides/speech-to-text
    model:
    # The language of the input audio. Supplying the input language in ISO-639-1 format
    # (https://en.wikipedia.org/wiki/List_of_ISO_639-1_codes) will improve accuracy and latency.
    language:

  groq:
    # API key for Groq. See https://console.groq.com/keys
    api_key:
    # See https://console.groq.com/docs/speech-to-text
    model:
    # The language of the input audio. Supplying the input language in ISO-639-1 format
    # (https://en.wikipedia.org/wiki/List_of_ISO_639-1_codes) will improve accuracy and latency.
    language:

  deepgram:
    # See https://console.deepgram.com/
    api_key:
    # See https://developers.deepgram.com/reference/speech-to-text-api/listen-streaming#request.query.model
    model:
    # See https://developers.deepgram.com/reference/speech-to-text-api/listen-streaming#request.query.language
    language:
    # Whether to return interim (non-final) transcription results. Defaults to true
    interim_results: true
    # Whether to apply smart formatting to numbers, dates, etc. Defaults to true
    smart_format: true
    # Whether to add punctuations to the transcription. Defaults to true. Turn detector will work better with punctuations
    punctuate: true
    # Whether to include filler words (um, uh, etc.) in transcription. Defaults to true
    filler_words: true
    # Whether to filter profanity from the transcription. Defaults to false
    profanity_filter: false
    # List of tuples containing keywords and their boost values for improved recognition. Each tuple should be (keyword: str, boost: float). Defaults to None. keywords does not work with Nova-3 models. Use keyterms instead.
    # keywords:
    #   - [OpenVidu, 1.5]
    #   - [WebRTC, 1]
    # List of key terms to improve recognition accuracy. Defaults to None. keyterms is supported by Nova-3 models.
    # Commented below is an example
    keyterms:
      # - "OpenVidu"
      # - "WebRTC"

  assemblyai:
    # API key for AssemblyAI. See https://assemblyai.com/app/
    api_key:
    # See https://www.assemblyai.com/docs/speech-to-text/streaming#add-custom-vocabulary
    # Commented below is an example
    word_boost:
      # - "OpenVidu"
      # - "WebRTC"
    # See https://www.assemblyai.com/docs/speech-to-text/streaming#disable-partial-transcripts
    disable_partial_transcripts: false

  fal:
    # API key for fal. See https://fal.ai/dashboard/keys
    api_key:
    # See https://fal.ai/models/fal-ai/wizper/api#schema
    task:
    # See https://fal.ai/models/fal-ai/wizper/api#schema
    language:
    # See https://fal.ai/models/fal-ai/wizper/api#schema
    chunk_level:
    # See https://fal.ai/models/fal-ai/wizper/api#schema
    version:

  clova:
    # Secret key issued when registering the app
    api_key:
    # API Gateway's unique invoke URL created in CLOVA Speech Domain.
    # See https://guide.ncloud-docs.com/docs/en/clovaspeech-domain#create-domain
    invoke_url:
    # See https://api.ncloud-docs.com/docs/en/ai-application-service-clovaspeech-longsentence
    language:
    # Value between 0 and 1 indicating the threshold for the confidence score of the transcribed text. Default is 0.5.
    # If the confidence score is lower than the threshold, the transcription event is not sent to the client.
    # For a definition of the confidence score see https://api.ncloud-docs.com/docs/en/ai-application-service-clovaspeech-grpc
    threshold:

  speechmatics:
    # API key for Speechmatics. See https://portal.speechmatics.com/manage-access/
    api_key:
    # See https://docs.speechmatics.com/rt-api-ref#transcription-config
    language:
    # See https://docs.speechmatics.com/features/accuracy-language-packs#output-locale
    output_locale:
    # See https://docs.speechmatics.com/features/realtime-latency#partial-transcripts
    enable_partials:
    # See https://docs.speechmatics.com/features/realtime-latency#configuration-example
    max_delay:
    # See https://docs.speechmatics.com/features/realtime-latency#configuration-example
    max_delay_mode:
    # See https://docs.speechmatics.com/features/punctuation-settings
    # Commented below is an example of punctuation settings
    punctuation_overrides:
      # permitted_marks: [ ".", "," ]
      # sensitivity: 0.4
    # See https://docs.speechmatics.com/features/custom-dictionary
    # Commented below is an example of a custom dictionary
    additional_vocab:
      # - content: financial crisis
      # - content: gnocchi
      #   sounds_like:
      #     - nyohki
      #     - nokey
      #     - nochi
      # - content: CEO
      #   sounds_like:
      #     - C.E.O.