English speech to text, built on PyAI Hear
English speech to text turns spoken English into written text your software can store, search and act on. PyAI Hear is the recognition engine behind it, the same one that powers live phone agents on PyAI. Hear reads clean studio audio and hard 8 kHz phone lines, handles accents from different English speaking regions, and returns text in plain, JSON or timestamped form. You call it from your own code and keep the audio in your own flow. Use it to transcribe English audio to text from a saved file, or as English voice to text for a clip you record right here on the page. As an audio to text generator, it suits one off files as well as steady pipelines. PyAI works as a free English speech to text generator, so a saved file or a quick recording turns into text you can edit, search and reuse.
Key takeaways
- Price: $0.001/min, billed by the whole minute.
- Free tests: $50 credit with no credit card, and sandbox keys never bill, so you can try English audio before paying anything.
- Accuracy: word error rate is 1.59% on clean English and 4.76% on real 8 kHz phone audio, PyAI's own benchmark numbers reported as industry parity rather than a lead.
- Speed: the first partial transcript lands in about 200 ms live, and long recordings run 92 to 247 times real time.
- Drop in: Hear is OpenAI compatible, so a whisper-1 client points at PyAI by changing the base URL, and single files up to 25 MB work in one call.
- Accents: US, UK, Indian and Australian English share one model.
How English transcription works
You send English audio to one endpoint and get text back. The same engine runs as a file call, a live stream, or a background job.
Post an English file to the transcription endpoint, or open a WebSocket and stream PCM16 frames for a live call or meeting.
The engine runs in one region, detects turns, and commits a final line once a speaker finishes, so partials firm up as more audio arrives.
Smart formatting, dictation and filler removal clean the text, and numerals are written the way an English reader expects.
Plain text, JSON, or verbose JSON with per segment timestamps and speaker turns, ready for notes, search or a knowledge base.
What Hear handles in English
English brings its own accents, numbers and names. Here is how Hear deals with them.
Regional accents
US, UK, Indian and Australian English share one model, so a mixed team on one call still lands in a single transcript without you picking a dialect up front.
Phone audio
Hear is tuned for 8 kHz phone lines, the hard case. Word error rate holds at 4.76% on noisy, accented call audio.
Numbers and names
Dates, money and order numbers are written in digits, and up to 5 custom vocabulary terms teach it the names and product words your business uses most.
Overlap and turns
A turn detector marks who spoke when, so overlapping speakers come back as separate labeled segments rather than one run on block.
Filler and dictation
Ums and false starts are removed, and spoken commands like new line are applied, so the transcript reads like written English.
Long recordings
A two hour interview is split into overlapping chunks, transcribed in parallel and stitched back, so throughput reaches 92 to 247 times real time.
Accuracy and speed
Hear is at industry parity for accuracy, held on the hard case of real phone audio. The figures below are PyAI's own measurements on English speech, not a per language claim.
What you can transcribe in English
The same English endpoint feeds notes, captions, search and analytics.
Call transcription
Log every English support or sales call as searchable text, then feed it to quality review, coaching or a CRM note.
Meeting notes
Stream a meeting live and get a running transcript with speaker labels, so action items are captured without a note taker.
Captions and subtitles
Generate English captions for video with per segment timestamps that drop straight into an SRT or VTT file.
Voice notes and dictation
Turn a quick voice memo into text in a note or ticket, with filler removed so it reads cleanly.
Interviews and podcasts
Transcribe long form English audio as a background job and get back timestamped text you can quote, clip or index.
Voice search and commands
Send short English clips and act on the text, so a user can speak a query or a command inside your app.
Formats, limits and output
| Input | Audio file up to 25 MB for a single file call, or a live PCM16 stream over WebSocket for real time. OpenAI compatible, so a whisper-1 client works unchanged. |
|---|---|
| Output | Plain text, JSON, or verbose JSON with per segment timestamps. Smart formatting, dictation and filler removal are applied on the way out. |
| Modes | Sync file call for short clips, live streaming for calls and meetings, and background jobs for long recordings that run 92 to 247 times real time. |
| Tuning | Set the language to English, add numerals handling, and load up to 5 custom vocabulary terms so names and product words land right. |
| Billing | $0.001/min of audio, rounded to the whole minute. An empty transcript costs nothing, and a busy engine returns a retry, not a charge. |
Call English speech to text from your own code
Hear speaks the OpenAI transcription format, so most code points at PyAI by changing the base URL. Full docs: PyAI SDKs.
SDK
From a server, the SDK posts an English file or opens a stream and returns the transcript in a few lines. The SDK speaks the whisper-1 shape, so existing transcription code moves over by swapping the base URL.
MCP
With the MCP server, an agent can transcribe an English recording on its own as a step in a flow, for example turning a call recording into notes, without you wiring the API by hand.
CLI
From the terminal, the CLI takes an English audio file or a folder and writes out text or timestamped JSON, which is handy for batch transcribing a backlog of recordings.
Where Hear fits
PyAI is voice infrastructure. Hear, Speak and Omni run as one stack in one region, so recognition, reasoning and speech share a session instead of hopping between vendors. Hear is not a consumer dictation app and PyAI does not resell another company's recognizer: it is the English transcription engine you call from your own product, on your own numbers or files. See the published numbers on PyAI benchmarks.
English speech to text FAQ
How accurate is English speech to text on Hear?
Hear measures 1.59% word error rate on clean English and 4.76% on real 8 kHz phone audio. Both figures are PyAI's own benchmark numbers and sit at industry parity, so they describe the engine overall rather than any single accent or recording.
How much does it cost to transcribe English audio?
Hear bills $0.001/min of audio, rounded to the whole minute, and an empty transcript is free. A new account starts with $50 of credit and no credit card, which covers a large amount of test audio before you spend anything.
Can I use my existing whisper-1 code?
Yes. Hear is OpenAI compatible, so a client written for whisper-1 works by pointing at the PyAI base URL. You keep your request shape and get text, JSON or verbose JSON with timestamps back the same way. Point the client at PyAI and the rest of your code stays put.
Does Hear handle live English audio or only files?
Both. A single file call suits short clips up to 25 MB, a WebSocket stream handles live calls and meetings with partials in about 200 ms, and a background job runs long recordings at 92 to 247 times real time.
Which English accents does it support?
US, UK, Indian and Australian English all run through one model, so you do not pick a dialect in advance. Custom vocabulary of up to 5 terms helps with names and product words that are specific to your audio. Test a few calls from your own callers to confirm the fit.
Prices checked November 2026. Hear figures from pyai.com/benchmarks.