Batch Transcription (Speech-to-Text)

Upload a complete audio file and receive a transcription in one batch request. Supports multiple languages via model selection. Optionally enables language correction and semantic tagging for enhanced output. For streaming results, use Real-time Transcription.

This endpoint is OpenAI SDK compatible. You can use the official OpenAI TypeScript or Python client libraries by setting the base URL to https://api.valsea.ai/v1. Authentication is required -- include your API key in the Authorization header or pass it via the SDK's apiKey parameter.

Form Fields

Parameter	Type	Required	Description
`file`	string (binary)	Yes	The audio file to transcribe (WAV, MP3, M4A, FLAC, OGG, WEBM). Maximum upload size: 10 MB. Maximum audio duration: 1 hour (3600 seconds).
`model`	`valsea-transcribe`	Yes	The transcription model to use. Always use `valsea-transcribe`. Language routing is determined by the `language` parameter.
`language`	string	Yes	Routing hint for the audio language, such as `english`, `singlish`, or `vietnamese`. See Language hints.
`response_format`	`json` \| `verbose_json`	No	Response verbosity level. Default: `json`
`enable_correction`	boolean	No	Enable language/grammar correction on the transcript. Default: `true`
`enable_tags`	boolean	No	Enable semantic tagging of the transcript.. Default: `true`
`diarize`	boolean	No	Enable speaker diarization. Requires `response_format=verbose_json`. Default: `false`
`diarization_min_speakers`	integer	No	Minimum expected speaker count for diarization. Default: `2`. Must be a positive integer (≥ 1).
`diarization_max_speakers`	integer	No	Maximum expected speaker count for diarization. Default: `6`. Must be ≥ 1 and ≥ `diarization_min_speakers` (no fixed upper limit).

Speaker Diarization

Set diarize=true to add speaker labels to the verbose response. Responses include speaker IDs on word objects (when words are returned) plus grouped utterances.

Speaker diarization is only available with response_format=verbose_json. Requests with diarize=true and response_format=json are rejected with 400 because minimal output cannot include speaker-level metadata.

Timestamp limitation (English): When diarize=true for English, word and utterance start / end timestamps will always be 0. Speaker labels and transcript text are still correct.

Language	Diarization	Speaker labels	Word / utterance timestamps
English	✅	✅	❌ always `0`
Vietnamese	✅	✅	❌ empty `words` array
Other languages	❌ not supported	—	—

Speaker diarization bills at 2x the normal transcription credit cost. If you also use paid rate-limit bypass, the request bills at 4x normal transcription credits.

The API playgrounds include a multi-speaker sample clip that automatically sets response_format=verbose_json, diarize=true, and a 2-speaker range for quick testing.

Paid Rate-Limit Bypass

If you need to exceed the transcription RPM limit for a request, you can bypass the rate-limit check by sending one of these opt-in flags:

Header: X-Bypass-Rate-Limit: true
Query parameter: bypass_rate_limit=true
Query parameter: bypassRateLimit=true

Bypass applies only to the per-organization rate-limit check. Authentication, credit checks, upload size limits, and concurrent upload limits still apply. Requests using this bypass are billed at 2x the normal transcription credit cost and include X-Credits-Multiplier: 2 in the response headers. If speaker diarization is also enabled, the multiplier becomes 4.

Code Examples

This endpoint is OpenAI SDK compatible. You can use the official OpenAI client libraries by pointing them at the VALSEA API base URL.

curl -X POST https://api.valsea.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "file=@audio.wav" \
  -F "model=valsea-transcribe" \
  -F "language=english" \
  -F "response_format=verbose_json" \
  -F "diarize=true"

Response

Minimal (`json`)

Field	Type	Description
`text`	string	The transcribed text.

Verbose (`verbose_json`)

Field	Type	Description
`text`	string	The transcribed text.
`raw_transcript`	string	Original transcript before corrections.
`detected_languages`	array	Languages detected in the audio.
`corrections`	array	List of corrections applied. Present only when corrections are generated.
`semantic_tags`	array	Semantic tags extracted from the text. Present only when semantic tags are generated.
`words`	array	Word-level timing and zero-based `speaker` labels when `diarize=true`.
`utterances`	array	Contiguous speaker turns with speaker, timing, transcript, and words when `diarize=true`.

Errors

Status	Description
`401`	Missing or invalid API key
`402`	Insufficient credits
`429`	Rate limit exceeded

Language Hints

The language field is a required routing hint for this batch endpoint. Provide the closest supported code—for example, english, singlish, vietnamese, or a regional variant—instead of testing separate endpoints.

Use a language hint such as english, singlish, vietnamese, arabic, arabic-egypt, or arabic-uae to select batch routing.

VALSEA supports 181 languages, dialects, and variants overall. Batch availability and feature coverage vary by language. See the current supported-language matrix for the authoritative list. For automatic detection, use Real-time Transcription.