The cheapest speech to text API, $0.0015 a minute
Compared on list prices with sources. Pay per second, 60 minutes free, no credit card.
The short answer
On the list prices we compared, Speakvora speech to text is the cheapest at $0.0015 per minute of audio ($0.09 per hour), billed per second, with 60 minutes free and no credit card. One request, one JSON response: send audio, get text.
Try it
Make a free key in the portal, then:
curl -X POST https://speakvora.com/api/v1/listen \
-H "x-api-key: YOUR_KEY" -H "content-type: application/json" \
-d "{\"audio_base64\": \"$(base64 < clip.wav | tr -d '\n')\", \"format\": \"wav\", \"language\": \"en\"}"Response:
{ "text": "Your refund will arrive in three to five business days.", "seconds": 3, "language": "en" }Silence or music comes back as empty text with a nonspeech field instead of invented words. Python, Node and curl examples are on the quickstart.
Price per minute, compared
| Provider and model | List price per minute |
|---|---|
| Speakvora speech to text | $0.0015 |
| ElevenLabs Scribe v2 ($0.22 per hour) | $0.0037 |
| OpenAI gpt-4o-mini-transcribe (estimated cost) | $0.0030 |
| Deepgram Nova-3, pre-recorded, monolingual (pay as you go) | $0.0043 |
| OpenAI whisper | $0.0060 |
List prices read on 8 November 2026; sources for every row, and more providers, are on the sourced pricing comparison. Deepgram also lists $0.0077 for streaming, and $0.0036 pre-recorded on its Growth plan with prepaid credits. Prices change: check each provider before you decide.
What you get
- Whisper large-v3-turbo, hosted, language detected automatically or set by you.
- Six formats: wav, mp3, m4a, ogg, flac, webm; up to 4.5 MB per request.
- Live mode: a WebSocket that transcribes speech as it happens and detects the end of an utterance.
- Pay as you go: per second of audio, no subscription and no monthly minimum.
Honest limits
- No word timestamps and no speaker labels yet.
- We have not published an accuracy benchmark; test on 100 of your own recordings before you commit.
- 4.5 MB per request: send long recordings in pieces ().
- The 60 free minutes are a one-time allowance, not a monthly plan.
- Speed figures are our own measurements, not a third-party benchmark.
More: pricing compared · free tier, no card · Whisper API alternative · transcribe audio guide · live transcription guide
Questions
What is the cheapest speech to text API?
On the list prices we compared (read 8 November 2026), Speakvora speech to text is the cheapest at $0.0015 per minute, which is $0.09 per hour of audio. OpenAI gpt-4o-mini-transcribe was about $0.0030 (an estimated cost), ElevenLabs Scribe v2 $0.0037, Deepgram Nova-3 pre-recorded $0.0043 and OpenAI whisper $0.0060 per minute.
Which model does it use?
OpenAI's open Whisper large-v3-turbo model, hosted for you. You do not run or tune anything: you send audio and get text.
Is there a free tier?
Yes. Every account gets 60 minutes of speech to text free, with no credit card. When the 60 minutes are used the API answers HTTP 402 free_allowance_used until you add a card; it never bills a card you did not add.
What audio formats can I send?
wav, mp3, m4a, ogg, flac and webm, from 1 KB up to 4.5 MB per request. For longer recordings, cut the audio into pieces and send them one after another.
Does it detect the language?
Yes. If you leave out the language it detects it automatically; you can also pass a language code such as en or es to pin it. Accuracy differs by language and we have not benchmarked each one.
Can it transcribe live speech?
Yes, there is a WebSocket for live transcription with automatic end-of-speech detection. In our production run on 8 October 2026 a 4 second clip took 0.67 s and a 15 second clip 0.90 s to come back (our own measurement, method on the pricing page).
Does it give word timestamps or speaker labels?
Not yet. The response is the text, the number of seconds of audio billed, and the language. If you need timestamps or speaker labels today, use a provider that offers them.
What are the limits?
Up to 4.5 MB of audio per request (send longer recordings in pieces), formats wav, mp3, m4a, ogg, flac and webm. No word timestamps and no speaker labels yet. We have not published an accuracy benchmark, so test on a sample of your own recordings. Audio is sent through our server to a hosted speech-recognition service to be transcribed.