Speech-to-text APIs
A speech-to-text API takes audio and hands back text with timestamps, and usually speaker labels. The split that matters is recorded files versus live streams: most vendors price them differently, and a voice agent needs the stream. Deepgram, AssemblyAI, Speechmatics and Gladia are speech companies first. ElevenLabs and OpenAI sell transcription next to their voice and language models. Rev adds human transcribers on the same account, Google Cloud bills it like any other cloud service, and Groq runs open Whisper models fast and cheap. 9 tools, in the order they appear in the directory. This is not a ranking.
Grouped by job: AI Agent Infrastructure
Groq
GroqCloud runs a short list of open models, including gpt-oss and Whisper for speech-to-text, on Groq's own LPU hardware behind an OpenAI-compatible API, with a free tier and a pay-as-you-go Developer plan.
Deepgram
Speech-to-text API: Nova-3 transcribes recorded files and live streams in 60+ languages, Flux adds turn detection for English voice agents, with diarization, keyterm prompting and redaction, billed per second. Text-to-speech too, and self-hosted containers on Enterprise.
AssemblyAI
Speech-to-text API: Universal-3.5 Pro transcribes recorded audio in 18 languages, Universal-2 in 99, and Universal-3.6 Pro Realtime streams in 32, with add-ons for diarization, PII redaction, translation and medical terms. An LLM gateway and text-to-speech sit on the same key.
ElevenLabs Scribe
ElevenLabs' speech-to-text models: Scribe v2 transcribes recorded audio in 90+ languages with keyterm prompting, Scribe v2 Realtime streams text back in about 150 ms, and a HIPAA-eligible medical model. Billed per hour of audio through the ElevenLabs API.
OpenAI Transcription
OpenAI's speech-to-text API: gpt-transcribe for recorded files with prompts, keywords and language hints, and gpt-live-transcribe for live audio over the Realtime API. whisper-1 and the gpt-4o transcribe models are deprecated and shut down on February 26, 2027.
Speechmatics
Speech-to-text API for recorded and live audio in 55+ languages: Melia 1 follows speakers who switch language mid-sentence, Linden 1 is tuned for voice agents and Oak 1 for clinical terms. Cloud in the EU, US or Australia, or on-prem and on-device on Enterprise.
Gladia
Speech-to-text API for recorded and real-time audio: Solaria-1 transcribes 100+ languages and Solaria-3 six at a lower rate, with diarization, custom vocabulary, translation and PII redaction. Prepaid credits in euros, no subscription.
Rev AI
Rev's speech-to-text API: Reverb models for English and 57 other languages, hosted Whisper Large, and human transcription on the same account, with async jobs, streaming over WebSocket or RTMP and forced alignment. Reverb's inference code is open source.
Google Cloud Speech-to-Text
Google Cloud's speech-to-text API: Chirp 3 and older models across a long list of languages and regions, plus telephony and medical models, with streaming, sync and batch recognition and a cheaper dynamic batch mode for audio that can wait.
What it is, where it runs, what an hour of audio costs.
Straight from each vendor's own pages. Plans move often, so the check date is below.
| Tool | What it is | Runs on | Open source | Price |
|---|---|---|---|---|
| Groq | Hosted Whisper large v3 and v3 turbo on Groq's LPU hardware: recorded files in, transcripts or English translations out, over an OpenAI-compatible API. | GroqCloud. Files up to 25 MB on the free tier, 100 MB on Developer. | Whisper weights are MIT (OpenAI). GroqCloud is not. | Per hour of audio: Whisper large v3 $0.111, large v3 turbo $0.04. 10-second minimum per request. Free tier with rate limits. |
| Deepgram | Speech-to-text API with Nova-3 for recorded files and live streams in 60+ languages, Flux for English voice agents with turn detection. Diarization, keyterms and redaction. Text-to-speech too. | Deepgram's cloud, with an EU endpoint. Self-hosted containers in your VPC or on-prem on Enterprise. | No. | $200 free credit, no card. Nova-3 per minute: $0.0043 recorded, $0.0077 streaming (English); $0.0052 and $0.0092 multilingual. Flux $0.0077. Billed per second. |
| AssemblyAI | Speech-to-text API: Universal-3.5 Pro for recorded audio in 18 languages, Universal-2 for 99, Universal-3.6 Pro Realtime streaming in 32. Add-ons for diarization, PII redaction, translation, medical terms. | AssemblyAI's cloud, US or EU at the same price. HIPAA BAA available. | No. | $50 free credit. Per hour recorded: Universal-3.5 Pro $0.21, Universal-2 $0.15. Streaming $0.15 (Universal-Streaming) or $0.45 (3.6 Pro Realtime), billed on session time. Add-ons extra. |
| ElevenLabs Scribe | ElevenLabs' transcription models: Scribe v2 for recorded audio in 90+ languages, Scribe v2 Realtime at about 150 ms, and a HIPAA-eligible medical model. | ElevenLabs' cloud, through the same API as its voices. | No. | Per hour of audio: Scribe v2 $0.22, Scribe v2 Realtime $0.39. Keyterm prompting +$0.05/hr (+$0.08 realtime). Medical $0.22/hr. |
| OpenAI Transcription | gpt-transcribe for recorded files with prompts, keywords and language hints; gpt-live-transcribe for live audio over the Realtime API. Older whisper-1 and gpt-4o transcribe models shut down February 26, 2027. | OpenAI's cloud. Files up to 25 MB per request. | No. The older Whisper model weights are MIT on GitHub. | Per minute: gpt-transcribe $0.0045, live transcription $0.017. Deprecated models: whisper-1 and gpt-4o-transcribe $0.006, gpt-4o-mini-transcribe $0.003. |
| Speechmatics | Speech-to-text for recorded and live audio in 55+ languages. Melia 1 follows mid-sentence language switches, Linden 1 is built for voice agents, Oak 1 for clinical terms. | Speechmatics' cloud in the EU, US or Australia. On-prem, containers and on-device on Enterprise. | No. | $100 free credit, no card. Per hour: Melia 1 $0.24 recorded, $0.40 live; Standard $0.45; Enhanced $0.75 recorded, $0.80 live; Linden 1 $0.30 live. 33% off if you allow training on your audio. |
| Gladia | Speech-to-text for recorded and real-time audio: Solaria-1 in 100+ languages, Solaria-3 in six at a lower rate. Diarization, custom vocabulary, translation, PII redaction. | Gladia's cloud. Custom hosting on Enterprise. Self-serve audio trains models unless you opt out. | No. | €50 free credit, prepaid. Per hour recorded: Solaria-3 €0.16, Solaria-1 €0.27. Real-time €0.68. Lower with volume. |
| Rev AI | Reverb speech-to-text for English and 57 other languages, hosted Whisper Large, and human transcription on one account. Async jobs, streaming over WebSocket or RTMP, forced alignment. | Rev's cloud. HIPAA processing once enabled on the account. | The API, no. Reverb's inference code is Apache-2.0 on GitHub. | Free credit for 5 hours. Per hour: Reverb English $0.20, other languages $0.30. Whisper Large $0.005/min. Human $1.99/min. 15-second minimum. |
| Google Cloud Speech-to-Text | Google Cloud's speech-to-text: Chirp 3 and older models across a long list of languages, telephony and medical models, streaming, sync and batch recognition. | Google Cloud, in global, US, EU and regional locations. Models vary by location. | No. | V2 Standard $0.016/min for the first 500,000 minutes a month, down to $0.004 past 2 million. Dynamic batch $0.003/min. New Google Cloud accounts get $300 credit. |
Checked 2026-11-01. Sources: Deepgram pricing, AssemblyAI pricing, AssemblyAI models, ElevenLabs API pricing, OpenAI pricing, OpenAI speech-to-text guide, Whisper on GitHub, Speechmatics pricing, Gladia pricing, Rev AI pricing, Rev AI streaming docs, Reverb on GitHub, Google Cloud Speech-to-Text pricing, Google Cloud supported languages, Groq speech-to-text docs.
Pick it if.
- DeepgramYou need live transcription by the minute at a low rate, and may want to self-host later.
- AssemblyAIYou want transcripts plus redaction, speaker labels and summaries from one vendor and one key.
- ElevenLabs ScribeYou already use ElevenLabs voices and want transcription on the same bill.
- OpenAI TranscriptionYou are already on the OpenAI API and want one more endpoint, not one more vendor.
- SpeechmaticsYour speakers switch languages mid-sentence, or the audio has to stay in a set region or on your own hardware.
- GladiaYou transcribe lots of recorded audio in many languages and want prepaid euro pricing.
- Rev AISome files need a human transcriber, and you want that on the same account as the API.
- Google Cloud Speech-to-TextYour stack already runs on Google Cloud and you want speech billed with everything else.
- GroqYou want open Whisper models on recorded files at the lowest price per hour here.
Each listing.
The directory page, alternatives in the same category, and a head-to-head where we have one.
- GroqAlternatives
- DeepgramAlternatives
- AssemblyAIAlternatives
- ElevenLabs ScribeAlternatives
- OpenAI TranscriptionAlternatives
- SpeechmaticsAlternatives
- GladiaAlternatives
- Rev AIAlternatives
- Google Cloud Speech-to-TextAlternatives
Category pages list every tool in that category, including ones left off this page. Built one of these? Get it featured in its category.
The 2 biggest AI stories, every morning.
One short email: the two stories that matter, and why.
You're in.
The Daily AI Brief will land in your inbox soon. If you don't see it, check Promotions or Spam and move it to your main inbox.