Quality on the actual call
Evaluate the audio your callers hear, the details the system captures and the action it completes. Native 8 kHz phone audio is available in Omni; use the supported audio format for each product.
PyAI Hear
Speech to text
Speech to text converts spoken audio into written words. Streaming recognition produces revisable partial text during a conversation; recorded transcription processes an uploaded recording.
Try the product with your own example
Open the existing product experience, choose the matching settings and evaluate your actual words or recording.
How Speech to text works
Speech to text converts spoken audio into written words. Streaming recognition produces revisable partial text during a conversation; recorded transcription processes an uploaded recording.
Turn customer calls into searchable transcripts, useful summaries and reviewable quality evidence. Use Hear when your application needs text from speech. Keep streaming partials separate from final transcripts, and use speaker diarization on supported recorded-transcription jobs when you need speaker turns.
How PyAI approaches speech to text
A clear integration path
Hear and Speak offer OpenAI-compatible speech endpoints. Omni uses the realtime PyAI SDK. Pair the conversation with AMD, Trace, Recap and managed telephony where your workflow needs them.
Predictable usage billing
Review the product meter, sandbox limits and managed telephony terms on the pricing page. Cloning enrollment and prompt-to-voice design are free; generated speech follows synthesis billing.
Published reference: ~200 ms Hear first partial, in-region. Source: Live in-region, streaming partials; benchmark snapshot 2026-06-24. These are in-region reference measurements, not a language-specific result or latency guarantee. Network, load and deployment affect what you observe. Read the measurement context.
How to use Speech to text with PyAI
- 01
Choose recorded transcription or a realtime Hear stream for your audio.
- 02
Select a supported language or automatic detection where available, and supply relevant vocabulary.
- 03
Use final transcript text in your workflow; review critical names, amounts and corrections against the audio.
Voices, accents and your own voice
Listen to stock voices before choosing the voice for your workflow. A language label does not promise every regional accent. For generated speech, you can also enroll a voice you have permission to use or design a voice from a prompt. Verify language and product compatibility before using the result.
Listen to the voice directoryPut it to work
Call review
Preserve the words, corrections and outcome a reviewer needs to understand what happened.
Post-call handoff
Turn the conversation into useful context for the next team member.
Compare on the workflow you need
Comparing PyAI with Deepgram? Use the same noisy phone recordings and score business details as well as word error rate. Measure first partial and final text separately.
Read the comparison and migration guidanceSpeech to text FAQ
What is Speech to text?
Speech to text converts spoken audio into written words. Streaming recognition produces revisable partial text during a conversation; recorded transcription processes an uploaded recording.
When should I use Speech to text?
Use Hear when your application needs text from speech. Keep streaming partials separate from final transcripts, and use speaker diarization on supported recorded-transcription jobs when you need speaker turns.
How do I get started with Speech to text?
Choose recorded transcription or a realtime Hear stream for your audio. Select a supported language or automatic detection where available, and supply relevant vocabulary. Use final transcript text in your workflow; review critical names, amounts and corrections against the audio.
How much does it cost, and can I test it first?
PyAI offers sandbox testing with limits. Live usage follows the product's billing terms; see https://pyai.com/pricing for current rates, inclusions and managed telephony charges. Free testing does not mean unlimited free production usage.
Can I keep my existing application?
Hear and Speak expose OpenAI-compatible speech APIs. Update your base URL and credentials, select supported language and voice options, and validate the response formats your application uses.
What should I test before using it for customer calls?
Test representative names, dates, amounts, audio conditions and corrections. For agents, include business-tool failures and handoff to a person; inspect the observed outcome, not just a transcript.
Can I use a cloned or designed voice?
Speak supports stock, cloned and prompt-designed voices. Cloning enrollment and voice design are free; synthesis follows the current billing terms. Use a voice you have permission to reproduce and verify its intended language and product compatibility.