Speech-to-Text API

The audio file to transcribe. You can either upload a file object to the API or provide a public URL to download the audio file. The upload size is limited to 100MB. When providing the audio file via URL the maximum file size is 1GB.

Supported audio and video file formats: mp3, wav, flac, aac, opus, ogg, m4a, mp4, mpeg, mov, webm, and more.

The format in which the generated transcript is returned. See example API responses above. Must be one of the following: json, text, srt, verbose_json, or vtt.

vtt and srt are file formats that can be used to add subtitles to videos. verbose_json contains additional information such as the duration of the audio and timestamps for each audio segment.

Set this parameter to true to enable speaker diarization. This will add speaker labels to the transcript.

When using speaker diarization you may also add the parameters min_speakers and max_speakers to your API call to improve the accuracy of the speaker labels. The min_speakers and max_speakers parameters specify the minimum and maximum number of speakers in the audio file.

Important: Make sure to set the response_format parameter to verbose_json to be able to access the speaker labels. You can find an example response above.

Note: As OpenAI doesn't support speaker diarization, you can't use the speaker_labels parameter with the OpenAI Python or Javascript library.

A text to guide the transcript's style or continue a previous audio transcript. The prompt should be in the same language as the audio.

Examples

• Prompts can be useful for fixing words or acronyms that the model might get wrong in the audio. For example, the following prompt improves the transcription of the words "NFT" and "DeFi": The transcript is about blockchain technology, including terms like NFTs and DeFi.. Alternately, the prompt can be a simple list of words: NFT, DeFi, DAO, DApp
• Sometimes the model skips punctuation in the transcript. You can avoid this by using a simple prompt with punctuation: Hello, welcome to the podcast.
• The model usually skips common filler words. If you want to keep the filler words in your transcript, use a prompt that includes them: Umm, let's see, hmm... Okay, here's what I'm, like, thinking.

The language of the input audio. If no language is provided we detect the language automatically. Supplying the input language can improve accuracy and latency.

Supported languages: english, chinese, german, spanish, russian, korean, french, japanese, portuguese, turkish, polish, catalan, dutch, arabic, swedish, italian, indonesian, hindi, finnish, vietnamese, hebrew, ukrainian, greek, malay, czech, romanian, danish, hungarian, tamil, norwegian, thai, urdu, croatian, bulgarian, lithuanian, latin, maori, malayalam, welsh, slovak, telugu, persian, latvian, bengali, serbian, azerbaijani, slovenian, kannada, estonian, macedonian, breton, basque, icelandic, armenian, nepali, mongolian, bosnian, kazakh, albanian, swahili, galician, marathi, punjabi, sinhala, khmer, shona, yoruba, somali, afrikaans, occitan, georgian, belarusian, tajik, sindhi, gujarati, amharic, yiddish, lao, uzbek, faroese, haitian creole, pashto, turkmen, nynorsk, maltese, sanskrit, luxembourgish, myanmar, tibetan, tagalog, malagasy, assamese, tatar, hawaiian, lingala, hausa, bashkir, javanese, sundanese, cantonese, burmese, valencian, flemish, haitian, letzeburgesch, pushto, panjabi, moldavian, moldovan, sinhalese, castilian, mandarin

A URL to which the API will send a POST request when the transcription is ready. The POST request will include the transcript in the specified response format.

The callback_url parameter is useful for long audio files that take a while to transcribe. Instead of waiting for the API to finish processing, you can provide a callback URL and the API will send the transcript to that URL when it's ready.

Note: As OpenAI doesn't support asynchronous requests, you can't use the callback_url parameter with the OpenAI Python or Javascript library.

Enable word-level timestamps by adding word to the array (eg timestamp_granularities[]=word). By default only timestamps for each segment are added to the response. To use this feature, response_format must be set to verbose_json.