Venice Transcription (/audio/transcriptions)
POST /api/v1/audio/transcriptions takes an audio file and returns text. It's OpenAI-compatible with multipart/form-data — the OpenAI SDK's audio.transcriptions.create() works unchanged.
Use when
- You need STT (speech-to-text) for voice notes, meetings, podcasts, short audio.
- You need timestamps for subtitles / chapters.
- You want to pick between fast local-style models (Parakeet) and large multilingual ones (Whisper, Wizper, Scribe).
For long video / YouTube transcription, see venice-video's /video/transcriptions (takes a public video URL directly).
Minimal request
curl https://api.venice.ai/api/v1/audio/transcriptions \
-H "Authorization: Bearer $VENICE_API_KEY" \
-F "file=@./meeting.m4a" \
-F "model=nvidia/parakeet-tdt-0.6b-v3" \
-F "response_format=json" \
-F "timestamps=false"
{ "text": "Alright everyone, let's kick off the meeting..." }
With timestamps=true, json format also returns segment/word timings (schema is model-specific).
Request (multipart/form-data)
| Field |
Type |
Default |
Notes |
file |
binary |
— |
Required. Audio file. Supported: wav, wave, flac, m4a, aac, mp4, mp3, ogg, webm. Base64 is not accepted — upload as a real file. |
model |
enum |
nvidia/parakeet-tdt-0.6b-v3 |
See models below. |
response_format |
json / text |
json |
text returns text/plain body. |
timestamps |
bool |
false |
Include segment/word timestamps (JSON only). |
language |
string |
— |
ISO 639-1 hint (e.g. en, ja). Only Whisper-family models honor it; others auto-detect. |
Models
| Model ID |
Notes |
nvidia/parakeet-tdt-0.6b-v3 |
Default. Fast, English-first, great for real-time-ish flows. |
openai/whisper-large-v3 |
Large multilingual, honors language hint. |
fal-ai/wizper |
Whisper variant, competitive on quality/latency tradeoff. |
elevenlabs/scribe-v2 |
ElevenLabs Scribe, strong on noisy audio. |
stt-xai-v1 |
xAI Speech-to-Text. |
GET /models?type=asr returns the current catalog. ASR pricing is pricing.per_audio_second.usd — cost scales with audio duration.
OpenAI SDK
import OpenAI from 'openai'
import fs from 'node:fs'
const client = new OpenAI({
apiKey: process.env.VENICE_API_KEY,
baseURL: 'https://api.venice.ai/api/v1',
})
const out = await client.audio.transcriptions.create({
file: fs.createReadStream('meeting.m4a'),
model: 'openai/whisper-large-v3',
response_format: 'json',
language: 'en',
// @ts-expect-error — Venice-specific extra, passes through multipart
timestamps: true,
})
console.log(out.text)
Batch / long files
Venice doesn't expose native chunking. For files > ~30 min, split client-side on silence with ffmpeg or pydub, transcribe each chunk, then concatenate with offset timestamps.
ffmpeg -i long.mp3 -f segment -segment_time 600 -c copy chunk_%03d.mp3
Errors
| Code |
Meaning |
400 |
Bad params, unsupported audio format, empty file, or file larger than 25 MB (this endpoint returns 400 with "Maximum size is 25MB", not 413). |
401 |
Auth / Pro-only. |
402 |
Insufficient balance. |
415 |
Wrong Content-Type — must be multipart/form-data. |
422 |
Validation / upstream ASR error (e.g. zero-length audio, upstream provider 422). Not a "content policy" code on this path. |
429 |
Rate limited. |
500 / 503 |
Transient; retry with jitter. |
Gotchas
file must be uploaded as a real multipart file part. JSON + base64 is not supported here.- Timestamps are only surfaced in the JSON response shapes (
json, verbose_json, srt, vtt). With response_format: text the handler returns a plain text/plain body containing just the transcript — you'll lose any timestamp data, so pick verbose_json / srt / vtt when you need timings.
language is Whisper-specific. Parakeet / Scribe ignore it and auto-detect.- Peak concurrency limits apply — on
429, back off; big batches should throttle to ~5 parallel requests.
- Content-policy rejection on the transcript is returned as
422 with an error string; it does not surface suggested_prompt on this path.
1---2name: venice-venice-audio-transcription3description: Transcribe audio files to text via POST /audio/transcriptions. Covers supported models (Parakeet, Whisper, Wizper, Scribe, xAI STT), supported formats (wav/flac/m4a/aac/mp4/mp3/ogg/webm), response formats (json/text), timestamps, and language hints. OpenAI-compatible multipart.4license: MIT5---6
7# Venice Transcription (`/audio/transcriptions`)
8
9`POST /api/v1/audio/transcriptions` takes an audio file and returns text. It's OpenAI-compatible with `multipart/form-data` — the OpenAI SDK's `audio.transcriptions.create()` works unchanged.
10
11## Use when
12
13- You need STT (speech-to-text) for voice notes, meetings, podcasts, short audio.
14- You need timestamps for subtitles / chapters.
15- You want to pick between fast local-style models (Parakeet) and large multilingual ones (Whisper, Wizper, Scribe).
16
17For long video / YouTube transcription, see [`venice-video`](../venice-video/SKILL.md)'s `/video/transcriptions` (takes a public video URL directly).
18
19## Minimal request
20
21```bash
22curl https://api.venice.ai/api/v1/audio/transcriptions \
23 -H "Authorization: Bearer $VENICE_API_KEY" \
24 -F "file=@./meeting.m4a" \
25 -F "model=nvidia/parakeet-tdt-0.6b-v3" \
26 -F "response_format=json" \
27 -F "timestamps=false"
28```
29
30```json
31{ "text": "Alright everyone, let's kick off the meeting..." }
32```
33
34With `timestamps=true`, `json` format also returns segment/word timings (schema is model-specific).
35
36## Request (`multipart/form-data`)
37
38| Field | Type | Default | Notes |
39|---|---|---|---|
40| `file` | binary | — | **Required.** Audio file. Supported: `wav`, `wave`, `flac`, `m4a`, `aac`, `mp4`, `mp3`, `ogg`, `webm`. Base64 is **not** accepted — upload as a real file. |
41| `model` | enum | `nvidia/parakeet-tdt-0.6b-v3` | See models below. |
42| `response_format` | `json` / `text` | `json` | `text` returns `text/plain` body. |
43| `timestamps` | bool | `false` | Include segment/word timestamps (JSON only). |
44| `language` | string | — | ISO 639-1 hint (e.g. `en`, `ja`). Only Whisper-family models honor it; others auto-detect. |
45
46## Models
47
48| Model ID | Notes |
49|---|---|
50| `nvidia/parakeet-tdt-0.6b-v3` | Default. Fast, English-first, great for real-time-ish flows. |
51| `openai/whisper-large-v3` | Large multilingual, honors `language` hint. |
52| `fal-ai/wizper` | Whisper variant, competitive on quality/latency tradeoff. |
53| `elevenlabs/scribe-v2` | ElevenLabs Scribe, strong on noisy audio. |
54| `stt-xai-v1` | xAI Speech-to-Text. |
55
56`GET /models?type=asr` returns the current catalog. ASR pricing is `pricing.per_audio_second.usd` — cost scales with audio duration.
57
58## OpenAI SDK
59
60```ts
61import OpenAI from 'openai'
62import fs from 'node:fs'
63
64const client = new OpenAI({
65 apiKey: process.env.VENICE_API_KEY,
66 baseURL: 'https://api.venice.ai/api/v1',
67})
68
69const out = await client.audio.transcriptions.create({
70 file: fs.createReadStream('meeting.m4a'),
71 model: 'openai/whisper-large-v3',
72 response_format: 'json',
73 language: 'en',
74 // @ts-expect-error — Venice-specific extra, passes through multipart
75 timestamps: true,
76})
77
78console.log(out.text)
79```
80
81## Batch / long files
82
83Venice doesn't expose native chunking. For files > ~30 min, split client-side on silence with `ffmpeg` or `pydub`, transcribe each chunk, then concatenate with offset timestamps.
84
85```bash
86ffmpeg -i long.mp3 -f segment -segment_time 600 -c copy chunk_%03d.mp3
87```
88
89## Errors
90
91| Code | Meaning |
92|---|---|
93| `400` | Bad params, unsupported audio format, empty file, or **file larger than 25 MB** (this endpoint returns `400` with `"Maximum size is 25MB"`, not `413`). |
94| `401` | Auth / Pro-only. |
95| `402` | Insufficient balance. |
96| `415` | Wrong `Content-Type` — must be `multipart/form-data`. |
97| `422` | Validation / upstream ASR error (e.g. zero-length audio, upstream provider 422). Not a "content policy" code on this path. |
98| `429` | Rate limited. |
99| `500` / `503` | Transient; retry with jitter. |
100
101## Gotchas
102
103- `file` must be uploaded as a real multipart file part. JSON + base64 is **not** supported here.
104- Timestamps are only surfaced in the JSON response shapes (`json`, `verbose_json`, `srt`, `vtt`). With `response_format: text` the handler returns a plain `text/plain` body containing just the transcript — you'll lose any timestamp data, so pick `verbose_json` / `srt` / `vtt` when you need timings.
105- `language` is Whisper-specific. Parakeet / Scribe ignore it and auto-detect.
106- Peak concurrency limits apply — on `429`, back off; big batches should throttle to ~5 parallel requests.
107- Content-policy rejection on the transcript is returned as `422` with an error string; it does not surface `suggested_prompt` on this path.
yanacuti1121