VoiceAgentInputTranscription interface

Optional customer custom speech deployment configuration, keyed by locale.

Controls how long the model waits before emitting transcription text. Higher values can improve transcription accuracy at the cost of latency. Only supported with gpt-realtime-whisper in GA Realtime sessions.

Words or phrases to guide transcription of the input audio. Supported by gpt-transcribe and gpt-live-transcribe.

The language of the input audio. Supplying the input language in ISO-639-1 (e.g. en) format will improve accuracy and latency. Optional for other transcription models; required and must be nonempty when model is azure-speech-streaming (such as en-US; comma-separated locales are supported).

Possible languages of the input audio, in ISO-639-1 format. Supported by gpt-transcribe and gpt-live-transcribe.

The transcription model identifier. Configure customer custom speech deployments in custom_speech. azure-speech-streaming is for cascaded text-model conversations only (not realtime speech-to-speech models), requires a nonempty language such as en-US (comma-separated locales are supported), and supports custom_speech, phrase_list, and profanity_filter.

Optional phrase hints that bias recognition toward domain terms.

Controls how profanity is handled in input transcripts. Supported only with managed Azure Speech (azure-speech and azure-speech-streaming), including Custom Speech. Set this property before the session starts. If omitted, the transcription provider's default applies. You cannot change the mode during a session. Filtering affects transcript events and the text sent to cascaded models. It does not change input audio, voice activity detection, or assistant speech; audio-native models still receive the original audio. Mandatory service masking policies apply even when you select unfiltered.

An optional text to guide the model's style or continue a previous audio segment. For whisper-1, the prompt is a list of keywords. For gpt-4o-transcribe models (excluding gpt-4o-transcribe-diarize), the prompt is a free text string, for example "expect words related to technology". Prompt is not supported with gpt-realtime-whisper in GA Realtime sessions.