Murf Speech to Text (Beta): $0.24 an Hour, Now for Indian, US/UK and European Languages
Disclosure: links to Murf on this page are referral links. If you sign up through them we may earn a commission, at no extra cost to you. Prices,.

Disclosure: links to Murf on this page are referral links. If you sign up through them we may earn a commission, at no extra cost to you. Prices, limits and features were read off Murf’s API pricing page, its API changelog and its Speech to Text docs on 30 October 2026. We have not run our own accuracy test yet.
Murf has always been the voice that speaks. On 22 October it added the ear: Murf speech to text, a live transcription API now in beta. It is aimed squarely at Indian voice agents and call centres, and at $0.24 an hour it undercuts the obvious rival on price. Here is what it does, what it costs, and what the beta is still missing.
The short version
What: real-time transcription over one WebSocket, with live partial transcripts and a final transcript with word timings at the end of each turn.
Languages: English (India) plus 11 Indian languages, with automatic detection.
Price: $0.24 per hour, billed per second. The $10 free trial credit covers about 41 hours.
Limits: two regions (India for Finch-Indic, US-East for Finch-Global, since 2 Nov); 5 parallel streams on the free tier, 20 by default on pay-as-you-go (up to 150 on request).
Missing in the beta: no automatic end-of-turn detection yet. (Background-voice removal went live on 2 November.)

bvc=true) is now live. The original model is now called Finch-Indic, and the WebSocket URL needs a model parameter. Price is unchanged at $0.24 an hour. More in Voice AI Weekly #1.What Murf speech to text does
You open one WebSocket connection per call, stream audio in, and get text back while the person is still talking. The features, from Murf’s streaming docs:
- Partial transcripts that update as the caller speaks. Each one holds the whole turn so far, so you replace the text on screen rather than adding to it.
- A final transcript per turn, with start and end times for every word, once your app signals the end of the turn.
- Multi-turn sessions: one connection can stay open for a whole call. You send
Finalizeafter each turn and keep streaming. - Telephony audio: 8 kHz µ-law and A-law, plus PCM and Opus up to 48 kHz.
- Keyword boosting for brand names and product terms, with a 0–1 strength setting.
- Number formatting (an optional setting) for Hindi and English (India) transcripts.
The 12 languages
English (India), Hindi, Bengali, Tamil, Telugu, Kannada, Malayalam, Marathi, Gujarati, Punjabi, Odia and Assamese. You can name one language, a short list to detect from, or let it detect automatically. Update, 2 November: a second model, Finch-Global, now covers English (US, UK, Australia), German, Spanish, French, Italian, Portuguese and Dutch.
Murf speech to text pricing: what $0.24 an hour buys
Speech to Text is billed by the second at a flat $0.24 an hour, from the same pay-as-you-go balance as Murf’s voice API. Purchased credit never expires. Here is what common volumes cost, next to ElevenLabs’ listed rate for its live transcription model, Scribe v2 Realtime:
| Audio per month | Murf Speech to Text ($0.24/hr) | ElevenLabs Scribe v2 Realtime ($0.39/hr list) |
|---|---|---|
| 41 hours | Free (covered by the $10 trial credit) | $16 |
| 100 hours | $24 | $39 |
| 500 hours | $120 | $195 |
| 1,000 hours | $240 | $390 |
Two caveats keep this fair. ElevenLabs sells hours inside monthly plans (its free tier includes 2.5 hours of Scribe v2 Realtime, Creator 56 hours), and says annual Business plans pay $0.28 an hour or less. It also covers 90+ languages and publishes a latency of about 150 ms; Murf has not published a latency figure. Where you need only Hindi, Tamil or another of Murf’s 12 languages, Murf is the cheaper list price. For anything else, it isn’t an option yet. Our Murf vs ElevenLabs pricing guide covers the voice side.
Parallel streams
| Plan | Speech to Text streams at once | Text to Speech streams at once |
|---|---|---|
| Free trial ($10 credit) | 5 | 5 |
| Pay as you go | 20 by default, up to 150 on request | 5 |
| Enterprise | Custom | Custom |
What the beta is still missing
- Two regions. Finch-Indic runs on
wss://in.api.murf.ai; since 2 November, Finch-Global runs on a US-East endpoint for English (US, UK, Australia) and six European languages. - No automatic end of turn. The server keeps sending partial transcripts through silence. Your own voice-activity detection has to send
Finalize. Thecommit_mode=autosetting is accepted but ignored for now. - Background-voice removal is now live (
bvc=true, added 2 November, off by default). - Server-side keys only. Speech to Text does not accept Murf’s short-lived auth tokens, so you can’t safely connect straight from a browser or phone app. Route audio through your own server.
- No published accuracy or latency numbers. Murf has not shared word-error rates or response times for the beta.
The other Murf API changes in October
Speech to Text was the headline, but Murf’s API changelog had a busy month. If you build on the API, the first and last rows below may need code changes.
| Date | Change | What to do |
|---|---|---|
| 5 Oct | Gen2 retired from the streaming API | Move streaming calls to Falcon. Gen2 still works for non-streaming synthesis. |
| 9 Oct | Word timestamps on WebSocket streaming (Falcon 2) | Useful for handling a caller who interrupts your agent mid-sentence. |
| 13 Oct | Volume offset (−50 to 50) on Falcon 2 | Level voices without post-processing. |
| 15 Oct | Exact pronunciation via inline IPA or a pronunciation dictionary | Fix brand names and Indian place names once. |
| 19 Oct | Falcon 2.1 launched on streaming | Murf says it sounds more natural, starts speaking sooner and has fewer glitches. |
| 22 Oct | Speech to Text beta, plus usage headers that show how many of your parallel streams are in use | Watch the headers before you hit your limit. |
| 27 Oct | Voice list now split by model (falcon-2.1 or gen2) | Fetch the matching list before sending a voice ID. A voice in one model’s list may not exist in the other. |
Prices for the voice side, from the same pricing page: Falcon 2.1 costs $0.01 per 1,000 characters, Gen2 $0.03, translation $0.02 per 1,000 characters, and the Voice Changer $0.10 a minute. Our full Murf AI pricing guide covers the Studio plans.
What we could not verify
Open questions, stated plainly
Accuracy. No word-error rates published, and we have not tested it ourselves yet.
Latency. No time-to-first-transcript figure is given for Speech to Text.
When the beta ends, and whether $0.24 an hour stays the price after it.
A date for automatic end-of-turn detection, still listed as coming soon.
Verdict
If you are building a Hindi, Tamil or other Indian-language voice agent, Murf speech to text is worth a test now. It is the cheapest list price we have seen for live transcription, it fits telephony audio, and you can already run the whole speak-and-listen loop on one Murf account. The free $10 credit buys about 41 hours of testing. Treat it as a beta, though: plan your own end-of-turn detection, keep audio on your server, and check accuracy on your real calls before switching. If you need languages outside India, it is not for you yet.
Get Murf API ($10 free credit) →
Sources, all read 30 October 2026: Murf API changelog (5–27 Oct entries), Murf Speech to Text streaming docs, Murf’s API pricing page, and ElevenLabs API pricing.
FAQ
How much does Murf speech to text cost?
$0.24 per hour of audio, billed per second, on Murf’s pay-as-you-go API. New accounts get $10 of free API credit, which covers about 41 hours of transcription.
Which languages does Murf speech to text support?
Two models. Finch-Indic: English (India), Hindi, Bengali, Tamil, Telugu, Kannada, Malayalam, Marathi, Gujarati, Punjabi, Odia and Assamese. Finch-Global (since 2 November 2026): English (US, UK, Australia), German, Spanish, French, Italian, Portuguese and Dutch. Both detect the language automatically.
Is Murf speech to text cheaper than ElevenLabs?
On list price, yes. Murf charges $0.24 an hour; ElevenLabs lists Scribe v2 Realtime at $0.39 an hour (and says $0.28 or lower on annual Business plans). ElevenLabs covers 90+ languages, though, so it only competes where Murf’s languages are enough.
Does Murf speech to text detect when the caller stops talking?
Not yet. In the beta a turn ends only when your app sends Finalize. An automatic end-of-turn mode is listed in the docs as coming soon and is ignored today.
Can I still stream Murf Gen2 voices?
No. Murf retired Gen2 from its streaming API on 5 October 2026. Gen2 voices still work for non-streaming synthesis; streaming now uses Falcon 2.1, which has its own voice list.



