OSTT

Appearance

OpenAI ​

OpenAI provides speech-to-text models through its audio transcription API. OSTT supports the newer GPT-4o transcription models, GPT-4o diarization, and the legacy hosted Whisper model.

API documentation: OpenAI create transcription

Models ​

Model IDNotes
openai/gpt-4o-transcribeHigher-quality GPT-4o transcription model. Good default when accuracy matters.
openai/gpt-4o-mini-transcribeFaster, lighter GPT-4o transcription model. Good for high-volume use.
openai/gpt-4o-transcribe-diarizeGPT-4o transcription with diarized JSON output and speaker-segment annotations.
openai/whisper-1Legacy hosted Whisper model. Broad compatibility.

Select A Model ​

bash
ostt model select openai/gpt-4o-transcribe

Per command:

bash
ostt transcribe meeting.mp3 -m openai/gpt-4o-transcribe

Params ​

Configure persistent params in ~/.config/ostt/ostt.toml:

toml
[openai.gpt-4o-transcribe.params]
language = "sv"
prompt = "Technical meeting about OSTT, Whisper, and Rust."
temperature = 0.0
include = ["logprobs"]

[openai.whisper-1.params]
response_format = "verbose_json"
timestamp_granularities = ["word"]

[openai.gpt-4o-transcribe-diarize.params]
response_format = "diarized_json"
chunking_strategy = "auto"
known_speaker_names = ["agent"]

Override for one invocation with --param key=value:

bash
ostt transcribe meeting.mp3 -m openai/gpt-4o-transcribe --param language=sv --param temperature=0
ostt transcribe meeting.wav -m openai/gpt-4o-transcribe-diarize --param chunking_strategy=auto

List supported params for any model:

bash
ostt model params openai/gpt-4o-transcribe --format json

OSTT always returns plain transcript text. Params that require a JSON response are supported only when the response still contains a top-level text field.

GPT-4o Params ​

ParamTypeDescription
languagestringOptional language hint, such as en or sv. Use when you know the source language.
promptstringContext text that guides spelling, terminology, names, and writing style. Saved ostt keyword terms are used as fallback only when prompt is not set.
temperaturenumberSampling temperature, 0.0 to 1.0. Lower values are more deterministic.
includestring listlogprobs returns token log probabilities from gpt-4o-transcribe and gpt-4o-mini-transcribe. OSTT still outputs only transcript text.

Diarization Params ​

Use openai/gpt-4o-transcribe-diarize for diarization. OSTT sends response_format = "diarized_json" by default for this model and returns the combined text field.

ParamTypeDescription
languagestringOptional language hint.
promptstringContext text that guides vocabulary and style.
temperaturenumberSampling temperature, 0.0 to 1.0.
response_formatstringSupported values in OSTT: json, diarized_json. Defaults to diarized_json.
chunking_strategystringSupported value: auto. Used by OpenAI's diarization flow.
known_speaker_namesstring listOptional known speaker labels, sent as repeated known_speaker_names[] form fields.
known_speaker_referencesstring listOptional data URL audio references, sent as repeated known_speaker_references[] form fields.

Whisper Params ​

whisper-1 supports timestamp metadata through verbose_json. OSTT parses the text field and does not expose word or segment metadata in command output.

ParamTypeDescription
languagestringOptional language hint.
promptstringContext text for names and vocabulary. Saved ostt keyword terms are used as fallback only when prompt is not set.
temperaturenumberSampling temperature, 0.0 to 1.0.
response_formatstringSupported values in OSTT: json, verbose_json. text, srt, and vtt are not exposed because OSTT's OpenAI path parses JSON.
timestamp_granularitiesstring listSupported values: word, segment. Requires response_format = "verbose_json"; OSTT sets verbose_json automatically if timestamps are set without an explicit response format.

Not Exposed ​

OSTT does not expose OpenAI stream=true because this provider path expects a single JSON response. It also does not expose non-JSON response formats (text, srt, vtt) because they would bypass OSTT's JSON parsing path.