Type Alias DictationConfig

DictationConfig: {
    channels?: number;
    keyterms_prompt?: string[];
    language_codes?: string[];
    llm_instruction?: string;
    sample_rate?: number;
    stt_prompt?: string;
}

Options for a dictation request.

language_codes, stt_prompt and keyterms_prompt shape the transcript; llm_instruction asks the server to run a follow-up LLM pass over it, and the rewrite comes back as DictationResponse.llm_response. sample_rate and channels are needed only for raw PCM audio — a WAV container carries them in its own header.

These are the only fields sent: the dictation API accepts this exact set, so there is no model, prompt, timestamps or conversation_context.

Type declaration

  • Optionalchannels?: number

    The channel count (1 for mono, 2 for stereo). Setting either this or sample_rate marks the audio as raw 16-bit PCM, and both are then required. Leave both unset for WAV, which carries the channel count in its own header.

  • Optionalkeyterms_prompt?: string[]

    Terms to bias the decoder towards. Whitespace is stripped and empty terms are dropped. Maximum 100 terms and 8000 characters in total — lists over either cap are rejected.

  • Optionallanguage_codes?: string[]

    ISO 639-1 codes for the language(s) of the audio — a single-element array (e.g. ["es"]) for monolingual audio, or several codes (e.g. ["en", "es"]) for multilingual audio. Unset leaves the language to the server's default.

  • Optionalllm_instruction?: string

    Instruction for a follow-up LLM pass over the transcript, e.g. "Format this as a SOAP note.". The rewritten text comes back as llm_response; the raw transcript stays in text. Maximum 2048 characters; longer instructions are rejected.

  • Optionalsample_rate?: number

    The source sample rate in Hz. Setting either this or channels marks the audio as raw 16-bit PCM, and both are then required. Leave both unset for WAV, which carries the rate in its own header.

  • Optionalstt_prompt?: string

    Context for the transcription: a description of what the audio is about, e.g. "A doctor dictating a patient visit note.". It describes the situation rather than instructing the model, and steers the decoder as it writes the transcript — where llm_instruction reshapes the transcript afterwards. Maximum 6000 characters; longer prompts are rejected.

Settings

Member Visibility