`whisper-large-v3`

Full Whisper Large V3, including translate to English.

Native waveform: Accepts common audio containers. Language auto or a Whisper code.. Live price and availability are on GET /v1/models/whisper-large-v3.

Set task to translate to return an English transcript.

Parameters

Field	Type	Description
`params.language` Default auto	string	ISO 639-1 code. auto detects the language. ['af', 'am', 'ar', 'as', 'az', 'ba', 'be', 'bg', 'bn', 'bo', 'br', 'bs', 'ca', 'cs', 'cy', 'da', 'de', 'el', 'en', 'es', 'et', 'eu', 'fa', 'fi', 'fo', 'fr', 'gl', 'gu', 'ha', 'haw', 'he', 'hi', 'hr', 'ht', 'hu', 'hy', 'id', 'is', 'it', 'ja', 'jw', 'ka', 'kk', 'km', 'kn', 'ko', 'la', 'lb', 'ln', 'lo', 'lt', 'lv', 'mg', 'mi', 'mk', 'ml', 'mn', 'mr', 'ms', 'mt', 'my', 'ne', 'nl', 'nn', 'no', 'oc', 'pa', 'pl', 'ps', 'pt', 'ro', 'ru', 'sa', 'sd', 'si', 'sk', 'sl', 'sn', 'so', 'sq', 'sr', 'su', 'sv', 'sw', 'ta', 'te', 'tg', 'th', 'tk', 'tl', 'tr', 'tt', 'uk', 'ur', 'uz', 'vi', 'yi', 'yo', 'yue', 'zh']
`params.task` Default transcribe	string	transcribe returns the audio in its own language. translate returns an English transcript.
`params.response_format` Default verbose_json	string	json returns the transcript in result.text. verbose_json also returns language, duration, segments, and words. A word score is included only when word_probability is true. When timestamp_granularity is word or segment and this field is omitted, the API sets verbose_json.
`params.timestamp_granularity` Default word	string	none, segment, or word. none omits timing. segment returns a time range per phrase. word returns a time range per word. Omitting the field, or sending null, uses word. word and segment fill response_format to verbose_json when that field is omitted.
`params.prompt`	string	Up to 1000 characters. Steers the transcript: context, style, or what the model should listen for.
`params.temperature`	number or array	0 to 1. One value, or a fallback ladder of at most five. Omit to use 0, 0.2, 0.4, 0.6, 0.8, 1.0.
`params.hotwords`	string	Up to 500 characters. Words to boost in the transcript, such as names or jargon. Separate with spaces.
`params.vad_filter` Default false	boolean	Skip quiet stretches. Helps long files with gaps. Leave off for dense speech.
`params.min_silence_duration_ms` Default 1000 when vad_filter is true	integer	How long a pause must last before VAD treats it as silence. 0 to 30000. Only used when vad_filter is true.
`params.condition_on_previous_text` Default true	boolean	Feed earlier text into the next window. Turn off if the model starts repeating itself.
`params.prompt_reset_on_temperature` Default 0.5	number	0 to 1. When temperature rises past this, the previous-text prompt is dropped.
`params.language_detection_threshold` Default 0.5	number	0 to 1. How sure auto-detect must be to lock a language. Lower accepts weaker guesses. Used when language is auto.
`params.prefix`	string	Up to 500 characters. Force the transcript to start with this text.
`params.repetition_penalty` Default 1	number	1 to 2. 1 is off. Below 1 is rejected. Above 1 punishes repeats. Try 1.1 to 1.3 if it loops.
`params.no_repeat_ngram_size` Default 0	integer	0 to 20. 0 is off. 3 blocks the same 3-word run from appearing twice.
`params.hallucination_silence_threshold`	number	Seconds of silence after which leftover words are dropped as likely invented. Omit unless you see phantom text in gaps. 0 to 120.
`params.no_speech_threshold` Default 0.6	number	0 to 1. Higher skips more windows as no speech.
`params.log_prob_threshold` Default -1	number	-10 to 0. Average token confidence floor. More negative keeps weaker guesses. Leave default unless real words are being dropped.
`params.max_initial_timestamp` Default 999999	number	Cap on the first word start, in seconds. Default 999999. A low value stamps late speech as starting at that cap.
`params.length_penalty` Default 1	number	0 to 2. Default 1. Neutral length preference.
`params.compression_ratio_threshold` Default 2.4	number	1 to 10. Default 2.4. Raise if dense speech is dropped. Lower if you see looped letters.
`params.multilingual` Default false	boolean	Allow the file to switch languages mid-clip. Leave off unless the recording actually changes language.
`params.max_new_tokens`	integer	1 to 448. Cap on tokens per window. Omit for the model default.
`params.chunk_length`	integer	1 to 30. Seconds per window. Omit for the model default.
`params.language_detection_segments` Default 1	integer	1 to 10. How many windows to use when detecting language.
`params.word_probability` Default false	boolean	Include a 0 to 1 score on each word. Off when omitted. Segment confidence is unchanged.

Example

curl https://api.pascalai.co/v1/speech-to-text \
  -H "Authorization: Bearer $PASCAL_API_KEY" \
  -H "Idempotency-Key: $(uuidgen)" \
  -F model=whisper-large-v3 \
  -F params='{"task":"transcribe","language":"en"}' \
  -F file=@input.wav

curl https://api.pascalai.co/v1/speech-to-text/$JOB_ID \
  -H "Authorization: Bearer $PASCAL_API_KEY"