← Back to models

Whisper

transcriberAudio

Turn speech recordings into text with Whisper large-v3-turbo. Antfly’s transcription API accepts base64-encoded audio and can return timestamps. The pull command selects Antfly’s Q8_0 GGUF package of this checkpoint.

antfly inference pull antflydb/whisper-large-v3-turbo:gguf:Q8_0@9bc5ac765fa72da3a47580a86d5e0762a3242de4
By openaiLicense: mitHugging Face source

Technical Specs

Architecture
Whisper large-v3-turbo
Q8_0GGUFmultilingual

About#

Whisper large-v3-turbo is a pruned version of Whisper large-v3, cut from 32 decoder layers to 4. It runs much faster than large-v3 with a small loss in quality, and covers 99 languages. Antfly's package quantizes it to Q8_0: about 830 MB against about 1.5 GB for the published F16 weights, with identical transcripts in Antfly's runtime tests.

It is the largest Whisper checkpoint on this site. Use it when transcript accuracy matters more than decode time. Whisper small is the smaller choice that stays accurate on meeting audio, and Whisper tiny suits clean dictation where speed comes first.

Capabilities#

  • Speech to text in 99 languages
  • Phrase timestamps, so search hits can point to the moment a passage was spoken
  • Recordings of any length, transcribed in windows cut at pauses
  • WAV, FLAC, MP3, AAC, Opus, and Vorbis, plus video files transcribed from their audio track

Where it fits in Antfly#

Whisper is the transcriber enrichment for audio. Add a transcriber to an asset enrichment on a field that holds recording URLs, and Antfly transcribes each recording once on write, stores the transcript with its timing, and chunks it for full-text and vector search next to your other documents.

See Index Recorded Calls and Podcasts for the table setup.

Usage#

Pull and start Antfly#

antfly inference pull antflydb/whisper-large-v3-turbo:gguf:Q8_0@9bc5ac765fa72da3a47580a86d5e0762a3242de4

antfly standalone

Transcribe audio#

Send an audio recording to the read endpoint. Set timestamps in audio_options to include timing information in the response.

curl http://localhost:8080/ai/v1/read \
-H 'Content-Type: application/json' \
-d '{
"model": "antflydb/whisper-large-v3-turbo:gguf:Q8_0@9bc5ac765fa72da3a47580a86d5e0762a3242de4",
"audio": "data:audio/wav;base64,<base64-encoded-recording>",
"audio_options": {"timestamps": true}
}'