Skip to navigation

Whisper Large v3 Turbo

Whisper Large v3 Turbo
OpenAI · Transcription
POST /v1/audio/transcriptions

Controlled self-hosted Whisper Large v3 Turbo transcription with multilingual ASR, translation, VAD, timestamps, subtitles, hotwords, and decoder controls exposed.

See Whisper Large v3 Turbo pricing and specs or try it in the Playground.

At a glance

FieldValue
Model idwhisper-large-v3-turbo
Model release date2024-10-01
Input modalitiesAudio
Output modalitiesText
Context window-
Weight precisionFP16
Featurestranscription, translation, multilingual, word_timestamps, hotwords, srt_vtt
Native inferenceYes
NewNo
Supported endpointsPOST /v1/audio/transcriptions
Alternate model idsopenai/whisper-large-v3-turbo

Pricing

ChargeSpecRate
Controlled transcriptionper minute of audio$0.042 $0.005

Example request

curl https://api.empiriolabs.ai/v1/audio/transcriptions \
-H "Authorization: Bearer $EMPIRIOLABS_API_KEY" \
-F model=whisper-large-v3-turbo \
-F file=@meeting.mp3

Parameters

ParameterTypeRequiredDefaultDescription
audio_urlstringno-URL of the audio file to transcribe. Mutually exclusive with audio_base64.
audio_base64stringno-Base64-encoded audio bytes. Mutually exclusive with audio_url.
audio_suffixstringno".audio"File extension hint (mp3, wav, m4a, etc.) when the audio source has no recognizable extension.
languagestringno-ISO 639-1 language code (en, es, fr, etc.). Leave blank for auto-detection.
taskenumno"transcribe"transcribe = same language, translate = translate to English. · Allowed: transcribe, translate
beam_sizeintegerno5Beam search width. Higher = more accurate but slower. · Range: 1 – 32
best_ofintegerno5Number of candidates to sample with temperature > 0. · Range: 1 – 32
patiencenumberno1.0Beam search patience factor. Higher = explore more candidates. · Range: 0.0 – 10.0
length_penaltynumberno1.0Penalty applied to longer transcripts. Negative encourages shorter output. · Range: -10.0 – 10.0
repetition_penaltynumberno1.0Penalty for repeating tokens. >1 reduces repetition. · Range: 0.1 – 5.0
no_repeat_ngram_sizeintegerno0Block any n-gram of this size from repeating in the output. · Range: 0 – 20
temperaturestringno"0,0.2,0.4,0.6,0.8,1"Sampling temperature. 0 = deterministic, higher = more variation.
compression_ratio_thresholdnumberno2.4Treat output with compression ratio above this as failed and retry.
log_prob_thresholdnumberno-1.0Treat segments with average log-prob below this as failed and retry.
no_speech_thresholdnumberno0.6Mark a segment as silent when no-speech probability exceeds this AND log-prob is below threshold.
condition_on_previous_textbooleannotrueUse prior transcript as conditioning for the next segment.
prompt_reset_on_temperaturenumberno0.5Reset the conditioning prompt when temperature falls back during retry. · Range: 0.0 – 1.0
initial_promptstringno-Initial text prompt to guide vocabulary and style.
prefixstringno-Text to prepend to the first segment’s transcript.
suppress_blankbooleannotrueSuppress empty outputs at the start of each segment.
suppress_tokensstringno"-1"Comma-separated token IDs to suppress during decoding.
without_timestampsbooleannofalseDecode without timestamp tokens. Segments still carry their start and end times.
word_timestampsbooleannofalseInclude per-word timestamps in the response.
prepend_punctuationsstringno-Punctuation characters to merge with the following word.
append_punctuationsstringno-Punctuation characters to merge with the preceding word.
max_initial_timestampnumberno1.0Cap the first segment’s start time to this many seconds. · Range: 0.0 – 30.0
multilingualbooleannofalseAllow language switching within a single audio file.
vad_filterbooleannotrueApply Silero VAD to remove silence before decoding.
vad_parametersobjectno-VAD configuration as JSON (threshold, min_speech_duration_ms, etc.).
max_new_tokensintegerno-Cap on decoded tokens per segment.
chunk_lengthintegerno-Length of each audio chunk in seconds before decoding.
clip_timestampsstringno"0"Only decode within these (start, end) second ranges. Format: “0.5,12.3,15.0,30.0”.
hallucination_silence_thresholdnumberno-Treat long silent sections above this many seconds as hallucinations and skip them.
hotwordsstringno-Comma-separated hotwords to bias decoding toward (proper nouns, jargon).
language_detection_thresholdnumberno0.5Confidence threshold for auto language detection.
language_detection_segmentsintegerno1Number of leading segments to use for language detection. · Range: 1 – 20
include_tokensbooleannofalseInclude raw token IDs alongside each word/segment.
response_formatenumno"verbose_json"json | verbose_json | text | srt | vtt. · Allowed: verbose_json, json, text, srt, vtt

Notes

Supports URL/base64 audio, language/task, beam and temperature fallback controls, VAD/chunking, hotwords, prompts, word timestamps, punctuation controls, token debug output, and JSON/text/SRT/VTT formats.


Machine-readable schema: GET https://api.empiriolabs.ai/v1/models/whisper-large-v3-turbo.