Gemini 3.5 Transcribe

A Google model for speech recognition, language detection, speaker diarization and word-level timestamps.

Ailoot · Speech-to-text POST /v1/audio/transcriptions gemini-3.5-transcribe

About the model

This description covers the model’s capabilities. Available in Ailoot inputs, outputs and options are listed for the selected method below.

Gemini 3.5 Transcribe converts an audio file into text. Alongside transcription, it supports language detection, identification of speakers, timestamps for individual words, a custom vocabulary and smart text preparation.

Languages and duration

More than 85 languages are supported, including Russian, as well as switching between languages within a recording. Language is detected automatically. An ordinary request processes audio of up to one hour. With speaker diarization or word-level timestamps enabled, the limit drops to 30 minutes.

Output includes the complete transcript and, for the relevant modes, individual word annotations. The model is intended for file processing; continuous real-time recognition belongs to the separate gemini-3.5-transcribe-live version.

Transcription modes

Verbatim transcription is used by default. Smart mode removes filler words, accounts for speakers' self-corrections and formats the text. This mode cannot be combined with speaker diarization or word-level timestamps.

Speaker diarization supports up to eight speakers. Identifying who spoke a line has experimental status for three or more speakers. Timestamps help connect the text to the recording, but enabling them may reduce recognition accuracy.

Terms and files

The custom term vocabulary holds up to 1000 entries; Google notes that it is usually better to limit it to around one hundred. This vocabulary is incompatible with speaker diarization and word-level timestamps. Mode selection therefore needs to account for what matters more: specialized terms or detailed speaker and time annotations.

Supported audio formats include WAV, MP3, AIFF, AAC, OGG, FLAC, MPEG, M4A, L16, Opus, A-law, mu-law and WebM. Duration limits and permitted mode combinations are published for file processing. The Transcribe family description also specifies 96K tokens of context and 32K tokens of text output; these family characteristics do not replace file limits in the selected method.

Context caching, code execution, file search, function calling, image creation and Thinking are not supported. The model specializes in obtaining text and annotations from audio, not generating new speech.

Ailoot · Speech-to-text POST /v1/audio/transcriptions gemini-3.5-transcribe

gemini-3.5-transcribe

Google /gemini-3.5-transcribe

Overview

A Google model for speech recognition, language detection, speaker diarization and word-level timestamps.

Accepted inputs: Audio
Results: Text

Available in Ailoot · inputs & methods

Ailoot · Speech-to-text POST /v1/audio/transcriptions

Input types: audio
Output types: text
Output format: json

Field	Type	Requirement	Values & limits	Default
Field`model`	Type`string`	RequirementRequired	Values & limits gemini-3.5-transcribe	Default—
Field`file`	Type`file`	RequirementRequired	Values & limits Maximum bytes: 67,108,864 File formats: wav, mp3, m4a, ogg, webm, flac Content type: multipart/form-data Maximum duration, seconds: 600	Default—
Field`response_format`	Type`string`	RequirementOptional	Values & limits json	Default—

Input material combinations

speech_to_text · Required: actual audio file

Method parameters and lifecycle

Parameter	Contract
Parameter`material_types`	Contractaudio
Parameter`required`	Contractnull
Parameter`optional_controls`	Contractnull

Lifecycle: synchronous_result_then_owned_library

GET /v1/media/tasks/{task_id} · actual returned task_id or media library record

GET /v1/media/assets/{asset_id}/content

GET /v1/media/assets

GET /v1/media/assets/{asset_id}

Method shapes and IDs are exact; supplier capability is separate from admission.

Pricing

Ailoot method rates

Ailoot Speech-to-textPOST /v1/audio/transcriptions

Input: $0.01 per minute or part thereof

Ailoot reserves an estimated amount before a request. After the result, it settles the final charge from measured usage. Streaming settlement uses the final usage record. How reservation and settlement work →

Example requests

Ailoot · Speech-to-text

Transcription example

set -eu
: "${AILOOT_API_KEY:?Set your Ailoot API key}"
: "${AILOOT_AUDIO_FILE:?Set your actual MP3 recording}"
export AILOOT_AUDIO_FILE
python3 - <<'PY_REQUEST'
import os
from pathlib import Path
f=Path(os.environ["AILOOT_AUDIO_FILE"])
if not f.is_file() or f.suffix.lower()!=".mp3":raise SystemExit("Choose your actual MP3 recording")
PY_REQUEST
curl --fail-with-body --silent --show-error https://api.ailoot.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $AILOOT_API_KEY" \
  --form-string model=gemini-3.5-transcribe -F "file=@$AILOOT_AUDIO_FILE" \
  -D response-headers.txt -o result.json
python3 - <<'PY_RESULT'
import json
from pathlib import Path
print(json.dumps(json.loads(Path("result.json").read_text()),ensure_ascii=False,indent=2))
PY_RESULT

Receive and save results

Ailoot · Speech-to-text

json

GET /v1/media/tasks/{task_id}
GET /v1/media/assets/{asset_id}/content
GET /v1/media/assets
GET /v1/media/assets/{asset_id}

Keep all completed results

Keep every complete returned response, file and content part.

Retain complete JSON, not only text; preserve every available segment/speaker/timestamp and actual task/asset identity.

If the original response needs checking

Retain original operation_id/X-Client-Request-ID for reconciliation without resubmission; no dedicated audio poll route.

Response fields

Field	Type
Field`text`	Typestring
Field`segments[]`	Typeobject
Field`task_id`	Typestring
Field`asset_url`	Typestring
Field`usage`	Typeobject

Errors and next steps

Response	Condition	Next step
Response400 · `invalid_request_error`	ConditionInvalid required fields, material representation or parameter value	Next stepCorrect the indicated fields before a new creation request.
Response`server_error`	ConditionThe returned response says the operation or result requires a check	Next stepKeep the original returned operation/task ID, request headers and response; continue supported reads of that original operation. Do not resubmit the creation request.

Aliases

Model aliases

Ailoot gemini-3.5-transcribe

Method path aliases

Use the method paths shown above.

Related caller IDs

gemini-3.5-transcribe

FAQ · Gemini 3.5 Transcribe

What inputs and results are available?: POST /v1/audio/transcriptions · gemini-3.5-transcribe: audio → text
Which fields does the request use?: POST /v1/audio/transcriptions · gemini-3.5-transcribe: model, file, response_format

Gemini 3.5 Transcribe