Skip to content
Speech to text, built for real audio

Italian speech to text for voice agents

PyAI Hear writes down spoken Italian while the speaker is still going, on a self-service kiosk, inside an app or through a phone line. A first partial lands in about 200 ms, which gives your assistant a head start on its answer.

~200 ms first partialMilan to Palermo accentsKiosk, app or web$0.001/min

Explore first. Get a sandbox key when you are ready to build.

PyAI in motionInteractive example
Sample: an 18 second Italian kiosk order at a café, transcribed live
0:00ClienteVorrei due cappuccini e un cornetto alla crema, da portare via.
0:05AssistenteDue cappuccini e un cornetto alla crema. Desidera altro?
0:09ClienteNo grazie, pago con il telefono.
0:13AssistenteIl totale è 5,40 euro. Avvicini il telefono al lettore.
Drop an audio file, or record a clip

Try it free: 3 transcriptions, up to 3 minutes each, no account needed. After that, create a free account with $50 in credit.

  1. 1Upload or recordA file or a mic clip, up to 3 minutes
  2. 2Hear transcribesSpeech recognition tuned for real audio
  3. 3Read it hereThe transcript shows right on this page

A sample of how Hear returns Italian voice agent transcripts, with per segment timestamps and speaker labels.

Live Italian transcription across apps, kiosks and devices

An Italian speech to text API inside a voice agent works under time pressure that batch transcription never faces. Someone is waiting at a kiosk, in a car or in front of a phone screen, so Hear streams partial text in about 200 ms. Live Italian transcription from Hear can feed assistant code you own, or PyAI Omni can listen, reason and speak inside one session. The error rates quoted here come from English measurements and are not an Italian claim, so gather Italian recordings from your own product and test them first. The same engine also works as a Italian speech to text generator for the transcripts your team keeps after each session.

Key takeaways

  • First partial: about 200 ms after Italian audio reaches Hear.
  • Accuracy: At industry parity on English benchmarks, Hear reaches 1.59% WER for clean speech and 4.76% for noisy 8 kHz audio, with no Italian score yet.
  • Regional sound: Milanese, Roman, Neapolitan and Sicilian speakers all differ, so pilot with users from the regions you serve.
  • Price: $0.001/min billed by the whole minute, no platform fee, and silence that yields an empty transcript is free.
  • Omni timing: end of turn spotted about 196 ms after speech stops, with voice to voice near 390 ms median.
  • Migration: whisper-1 compatible, so current code switches by changing the base URL.

How Italian speech turns into text as people talk

Real time Italian speech to text lets Hear read audio during the sentence, so a kiosk or app reacts before the speaker pauses.

Pick up the voice

A user speaks to a kiosk, a phone app, a browser tab or a support line, and your client captures PCM16 audio.

Open a WebSocket

Your client streams frames of roughly 20 ms over one WebSocket per session, with Italian set as the language.

Receive live partials

Hear sends the first Italian partial in about 200 ms and corrects words as more audio comes in.

Run the next action

Your assistant, or Omni with its turn detector, uses the final Italian text to add an item, answer or hand off.

What Hear handles in live Italian speech

Spoken Italian is quick, full of elisions and shaped by strong regional accents. Each trait shows up in a Turin café, a Bari supermarket or a Rome hotel lobby alike.

Text before the pause

Italian speakers often run several clauses together without a break. Hear streams partial text in about 200 ms so the screen can show progress, while your logic waits for final text before committing an order.

Elision and fast speech

Forms such as l'ho, dell'ordine and c'è run words into each other, so test sample sessions to confirm intent matching still picks up the key terms.

Double consonants

Pairs like pala and palla or nono and nonno differ by one held consonant, so check them on your own Italian audio and add vital words as custom vocabulary.

Accents from north to south

PyAI publishes no regional Italian score. Record users from Milan, Rome, Naples and Palermo through a sandbox key, which never bills, and compare the transcripts.

Noise in bars and shops

Espresso machines, music and street noise sit under the voice at a counter. The 4.76% figure for noisy and accented audio was measured on English, so test with recordings from the actual site.

Numbers, menus and turns

With numerals handling on, prices and quantities return as digits, and up to 5 custom vocabulary terms cover menu or product names. Verbose JSON adds speaker turns with timestamps.

Accuracy and speed

Hear's word error rates are PyAI's own benchmarks, measured on English at industry parity. PyAI claims no Italian error rate. The Omni numbers reflect a whole Omni session from start to finish.

1.59%Word error rate, clean audio
4.76%Word error rate, 8 kHz phone and noisy
~200 msTo the first partial transcript
92 to 247xReal time, batch throughput

Where Italian live transcription fits a voice agent

From Trento to Catania, Italian users speak to assistants at counters, in apps and on the road. Italian voice assistant speech to text serves very different settings, and these six jobs show how wide they range.

Kiosk and café ordering

Let customers order coffee, pastries or a pizza by voice at a self-service kiosk and see the order build on screen.

Voice tutor app

Help learners practise Italian conversation, with timestamps that show where a pause or a hesitation happened.

Hotel and museum website help

Put a microphone on a booking or ticket site so visitors can ask about times and prices without typing.

Hands-free assistant on the road

Take spoken destinations and messages from drivers in Italian while their hands stay on the wheel.

Accessibility dictation

Give people who cannot type a way to write messages, forms and notes in Italian by speaking.

Smart speaker at home

Turn Italian commands into actions for lights, heating and shopping lists on a connected device.

Formats, limits and output

What an Italian live transcription request sends and receives
InputItalian speech streams in as PCM16 frames over a WebSocket, or uploads as a single file up to 25 MB. A whisper-1 client works once the base URL points at PyAI.
OutputItalian partial transcripts during speech, then final text. Pick text, JSON or verbose JSON with timestamps and speaker turns.
ModesSync file call for short clips, live streaming for apps, devices and voice sessions, and background jobs for long recordings that run 92 to 247 times real time.
TuningSet the language to Italian, enable numerals handling for prices and quantities, and add up to 5 custom vocabulary terms for product or place names.
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.

Connect Italian speech to text to your assistant

Hear uses the OpenAI transcription format, so an Italian assistant stack normally edits only the base URL. Guides sit at PyAI SDKs.

SDK

Inside your backend, the SDK starts one stream per Italian session and passes partial and final text to your code as Hear emits it.

MCP

With the MCP server, an agent framework calls Hear as a tool, for instance to transcribe an Italian voice memo before picking a next step.

CLI

From a terminal, the CLI streams an Italian sample and prints partials live, handy for checking menu terms before rollout.

Where Hear fits

PyAI is voice infrastructure. Hear, Speak and Omni share one stack in one region, so recognition, turn detection, reasoning and speech stay in one session for an Italian assistant. Hear is not a conversation designer and does not script your agent's answers. Run Hear inside an agent you already have, or choose Omni for the whole exchange. Numbers sit on PyAI benchmarks.

Get a sandbox key → $50 free credit. No credit card. Sandbox keys never bill.

Italian speech to text for voice agents FAQ

How quickly do Italian partials come back from Hear?

Hear delivers the first Italian partial roughly 200 ms after audio arrives and keeps updating it as the speaker continues. Inside a PyAI Omni session, the turn detector marks the end of a turn about 196 ms after speech stops, and voice to voice response time sits near 390 ms median.

Does Hear publish an Italian accuracy score?

No. The published results are 1.59% word error rate on clean audio and 4.76% on 8 kHz phone, accented and noisy audio, both measured on English by PyAI at industry parity. Italian results are not published, so test recordings from your own kiosk, app or device before going live.

Can Hear run inside my own Italian assistant?

Yes. Hear is whisper-1 compatible, so existing code normally needs only a new base URL. Stream PCM16 frames of about 20 ms over a WebSocket with Italian selected, then route partial and final text into the assistant you already operate on a kiosk, website or mobile app.

How much does Italian live transcription cost?

Hear bills $0.001 per minute of Italian audio, rounded to the whole minute, with no platform fee, and a session that returns an empty transcript is free. Every new account gets $50 in credit with no credit card required, and sandbox keys never bill during testing.

When should I choose Omni instead of Hear?

Choose Hear when your assistant already handles reasoning and speech output and only needs Italian text. Choose Omni to run listening, turn detection, reasoning and speaking together, with end of turn around 196 ms, voice to voice near 390 ms median, and a detector that holds while a speaker reads digits.

Prices checked November 2026. Hear figures from pyai.com/benchmarks.

Start transcribing with PyAI Hear

Spin up a sandbox key and turn calls, meetings and recordings into clean text in minutes. No credit card, and $50 in free credit.

Get a sandbox key →