Skip to content
Speech to text, built for real audio

Spanish speech to text for voice agents

A Spanish speaking customer at a self-service kiosk, in a mobile app or on a support line wants the assistant to keep up. PyAI Hear delivers real time Spanish speech to text, with the first partial near 200 ms, so ordering or searching starts while the person is still mid sentence.

Spain and Latin AmericaKiosk, app, web or lineWebSocket streaming$0.001/min

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

PyAI in motionInteractive example
Sample: a 14 second Spanish voice order at a restaurant kiosk
0:00AsistenteHola, ¿qué te gustaría pedir hoy?
0:02UsuarioDos tacos al pastor y una horchata grande, por favor.
0:06AsistentePerfecto. Son 185 pesos. ¿Para comer aquí o para llevar?
0:12UsuarioPara llevar, gracias.
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 Spanish voice agent transcripts, with per segment timestamps and speaker labels.

Live Spanish transcription for every channel

Spanish speech to text for voice agents has to keep pace with a person who is standing at a screen, holding a phone or driving. Hear accepts PCM16 audio in slices of about 20 ms over a live WebSocket and sends back the first partial roughly 200 ms after speech starts. The source can be a kiosk microphone, a web widget, an app or a support line. PyAI Omni, the speech to speech agent, runs on Hear too, so you can keep your own agent logic or let Omni listen, reason and reply in one session. Live Spanish transcription from the same stream also feeds captions. The same engine also works as a Spanish speech to text generator for the transcripts your team keeps after each session.

Key takeaways

  • Opening text: the first Spanish partial lands near 200 ms after a user starts.
  • Accuracy, English measured: 1.59% WER clean, 4.76% on 8 kHz phone, accented or noisy audio. PyAI benchmarks at industry parity, with no Spanish score published.
  • Channels: kiosks, apps, browser widgets, cars and lines all stream to one WebSocket endpoint.
  • Billing: $0.001/min, rounded to the whole minute, with no platform fee.
  • With Omni: turn end about 196 ms after silence, voice to voice about 390 ms median.
  • Migration: whisper-1 compatible, so swapping the base URL keeps your client.

From spoken Spanish to an action

Whether a user orders at a kiosk or asks an app a question, the path from voice to words follows the same four steps.

Someone speaks

A customer talks to a kiosk, a mobile app, a site widget or a line, in any Spanish variety.

Frames stream

Your client forwards PCM16 chunks of about 20 ms over a WebSocket as the words come out.

Text shows up

A first partial, such as dos tacos al pastor, appears near 200 ms and sharpens with each word.

The assistant responds

Your agent or Omni adds the item, runs the search or answers the question from the final text.

Spanish speech conditions Hear deals with

Real Spanish input mixes regional accents, borrowed words and spoken numbers, in quiet rooms and loud ones alike.

Variety across regions

Castilian, Mexican, Rioplatense and Caribbean speakers sound very different. Check recordings from the regions you serve, since PyAI publishes no Spanish accuracy score.

Vos and local words

Rioplatense vos forms and words such as celular or móvil should come back as spoken. Confirm that on your own samples before launch.

Spoken amounts and codes

Prices like ciento ochenta y cinco, postal codes and order numbers return as digits with numerals handling, ready for a cart or a form.

English mixed in

Spanish speakers drop in English brand and tech words. Up to 5 custom vocabulary terms keep those names spelled the way your menu or catalog lists them.

Loud places

Food courts, shops and cars add noise. Hear measures 4.76% WER on noisy and accented audio, a figure taken on English.

Early text and turns

Partials start before the user finishes, and verbose JSON returns speaker turns with timestamps for order logs or captions.

Accuracy and speed

PyAI publishes accuracy measured on English at industry parity. Treat the figures as a reference for Spanish, then review a batch of your own recordings. Omni timings come from complete Omni sessions.

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

Spanish jobs for a listening assistant

Spanish voice assistant speech to text fits any product where Spanish speakers would rather talk than tap. Six places, each with equal weight:

Self-service kiosks

Take a food or ticket order at a kiosk in Mexico City or Madrid by voice and fill the cart from the transcript.

Shopping app search

Let a buyer in Bogotá say what they want and turn the transcript into a product search inside the app.

Language practice apps

Hear a learner speak Spanish, compare the transcript with the target phrase and give feedback on the spot.

Warehouse voice logging

Let pickers call out bin and item numbers hands free, logging each one as digits in the inventory system.

Healthcare intake tablets

Collect a patient's symptoms and details by voice on a clinic tablet before the consultation starts.

Customer support phone line

Answer a Spanish support number, understand what the caller needs and route or resolve it while they wait.

Formats, limits and output

What a Spanish kiosk session sends to Hear and gets back
InputSpanish audio as PCM16 over a live WebSocket from a kiosk, app, browser or line, or a single file up to 25 MB. A whisper-1 client connects without code changes.
OutputStreaming partials while the user talks, then a final transcript. Choose text, JSON or verbose JSON carrying 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.
TuningPick Spanish as the language, switch on numerals handling so prices and codes come out as digits, and add up to 5 custom vocabulary terms such as menu or brand 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 Hear to a Spanish assistant

Hear speaks the OpenAI transcription format, so a Spanish assistant using the whisper-1 compatibility alias moves over with a base URL change. Reference: PyAI SDKs.

SDK

Inside a Spanish kiosk or app backend, the SDK keeps one stream per session and forwards each partial to your order or search logic as soon as Hear sends it.

MCP

Via the MCP server, an agent can call Hear as a tool to transcribe a recorded Spanish voice note before deciding how to respond.

CLI

Point the CLI at a Spanish recording to watch partials print live, handy for checking menu or product names before release.

Where Hear fits

PyAI is voice infrastructure. Hear, Speak and Omni run together in one region, so a Spanish session stays in one place from the first spoken syllable to the reply. Hear is not a script writer and does not plan what your assistant says. Figures on PyAI benchmarks come from English tests, so pilot with real Spanish speakers first.

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

Spanish voice agent questions

How fast does Hear transcribe spoken Spanish?

The first Spanish partial reaches your code about 200 ms after the user begins, and Hear keeps refining it as the sentence goes on. Under Omni, the turn closes around 196 ms after the person stops talking, and a spoken reply arrives near 390 ms median, quick enough for an order at a kiosk.

Is there a published Spanish accuracy score?

No. Hear's published numbers, 1.59% on clean audio and 4.76% on 8 kHz phone, accented or noisy audio, come from English testing at industry parity. A Spanish product team should transcribe real sessions from its own markets, such as Mexico, Spain or Colombia, and review the output by hand.

Will Hear work with a Spanish assistant I already built?

Yes. Your backend opens a WebSocket per session, streams PCM16 audio from the kiosk, app or browser, and sends Hear's partial and final text to your own logic and voice. Teams already using whisper-1 change the base URL, and output comes as text, JSON or verbose JSON.

How much does Spanish transcription on Hear cost?

Hear charges $0.001 per minute of audio, rounded to the whole minute, with no platform fee, and an empty transcript is free. New accounts get $50 in credit without a credit card, and sandbox keys never bill, so you can try a complete Spanish ordering flow first.

Do I need Hear or Omni for a Spanish assistant?

Choose Hear if you already run your own reasoning and voice and only need Spanish text fast. Choose Omni if you want PyAI to handle listening, turn detection, reasoning and speech in one session, with a turn detector that waits through unfinished thoughts and spoken digit strings.

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 →