What is a speech-to-text API?

A speech-to-text API transcribes spoken audio into written text. You upload a recording or pass its URL, or stream live audio, and the API returns the transcript, with timestamps, speaker labels or subtitle files on the models that support them.

Model types

What a speech-to-text API does

A speech-to-text API, also called a transcription or automatic speech recognition (ASR) API, converts the spoken words in an audio or video file into text. The simplest result is a plain transcript. Depending on the model, the API can also return:

  • Timestamps for each segment or word, for captions and searchable playback.
  • Speaker labels that mark who said each part of a conversation, a feature called diarization.
  • Subtitle files in SRT or VTT format.
  • Formatting of punctuation, numbers and dates, and masking of personal information.
  • Translation of the speech into English, or into other languages on models built for subtitling.

Many models detect the spoken language automatically, and some accept a list of key terms, such as product names or jargon, to recognize them more reliably.

File transcription and live transcription

For a recording, send POST /v1/audio/transcriptions with the model ID and either the file as a multipart upload or its HTTPS URL in file_url. The API returns the transcript in the response. A long transcription can return a job ID instead, which you poll at GET /v1/jobs/{job_id}, and the header Prefer: respond-async asks for a job ID at once.

For speech as it happens, such as a call or a live event, live transcription models stream text back over the realtime voice WebSocket at /v1/realtime. Depending on the model, partial words arrive while the person speaks, and a settled transcript arrives at each pause.

What speech-to-text is used for

  • Meetings and calls. Searchable notes, and summaries when you pass the transcript to an LLM API.
  • Captions and subtitles. Timed text for videos, courses and live streams.
  • Voice input. Dictation, voice messages and spoken commands in apps.
  • Media archives. Transcripts that make podcasts and videos searchable.
  • Localization. Transcripts and translated subtitle tracks that prepare a video for other languages.

What to compare between transcription models

  • Accuracy on your audio. Test with real recordings that include your accents, background noise, crosstalk and domain vocabulary.
  • Languages and dialects. Check every language you need, including speech that mixes languages.
  • Output detail. Word timestamps, speaker labels, subtitle formats and formatting options.
  • File or live. Live captions need a model that streams over the WebSocket. Finished recordings go to the file endpoint.
  • Price. Most transcription models bill by the length of the audio, per minute or per hour, and some bill by audio tokens, so cost grows with the length of the recording.

Side-by-side model comparisons show two models' capabilities and rates together.

Try and call a transcription model on EmpirioLabs AI

Browse the transcription models and upload a recording in the Playground to test one. To call a model, create an API key in the dashboard and send the audio to https://api.empiriolabs.ai/v1/audio/transcriptions. The transcription reference lists the request fields and response formats, and the Realtime Voice API guide covers live transcription.

Pay as you go, or choose a plan with weekly allowances. Current rates are on the pricing page.

Transcription models on EmpirioLabs AI

13 models, cheapest first where they share a billing unit. Each model's page has its full rate card, parameters and code.

Endpoints

POST /v1/audio/transcriptions WEBSOCKET /v1/realtime

Priced per minute or hour of audio

ModelLowest rateApplies to
StepAudio 2.5 ASR StepFun$0.05 per hour of audioTranscription
Qwen Audio 3.0 ASR Alibaba Cloud$0.252 per hour of audioTranscription
Whisper Large v3 Turbo OpenAI$0.042$0.005 per minute of audioControlled transcription
StepAudio 2.5 ASR Stream StepFun$0.40 per hour of audioTranscription
StepAudio 3 ASR Max StepFun$0.60 per hour of audioTranscription
Qwen Audio 3.0 ASR Stream Alibaba Cloud$0.648 per hour of audioTranscription
Deepgram Nova-3 Deepgram$0.014 per minute of audioTranscription
Deepgram Nova-3 Stream Deepgram$0.025 per minute of audioTranscription
OpenAI Whisper 1 OpenAI$0.030 per minutePer Minute of Audio

Also available: Sonilo Proofread, Qwen Audio 3.1 ASR, Qwen Audio 3.1 ASR Stream, Qwen Audio 3.1 ASR Message

Frequently asked questions

Is speech-to-text the same as ASR?

Yes. Automatic speech recognition (ASR) is the technical name for converting speech into text, and a speech-to-text or transcription API makes it available to your code.

Can a speech-to-text API tell different speakers apart?

Yes, on models that support diarization. The transcript labels each part with a speaker, so you can see who said what.

Can I transcribe live audio?

Yes. Live transcription models stream text back over a WebSocket while the person is speaking. Recordings use the file transcription endpoint instead.

Can I get subtitles from a transcription API?

Yes, on models that return SRT or VTT. These subtitle formats carry each line of text with its start and end time.

How is speech-to-text priced?

Most models bill by the length of the audio, per minute or per hour, and some bill by audio tokens. On EmpirioLabs AI you can pay as you go or choose a plan with weekly allowances.

Which Transcription model costs the least on EmpirioLabs AI?

StepAudio 2.5 ASR lists the lowest rate per minute of audio of the 9 Transcription models priced that way, followed by Qwen Audio 3.0 ASR, Whisper Large v3 Turbo. The price table on this page shows each model's lowest rate and what it applies to.

Ready to use better endpoints?

Check out our pricing or reach out if you want your own model deployed on our stack.