# Mihu Voice — Speech-to-Text and Text-to-Speech for Live Calls | Mihu AI

Mihu Voice · New

# Voices that sound human. Transcripts as people speak.

Mihu Voice is our own speech stack, built for live calls: Umay turns speech into text while the caller is still talking, and Ses turns text into natural speech in seven languages. One API, hosted in Europe or Türkiye, or on your own servers.

Try the voices (https://voice.mihu.ai/#voices) Get an API key (https://voice.mihu.ai/console/keys)

~80 ms

Final text after the caller stops

7

Languages, speech-to-text and text-to-speech

10–30 s

Of speech to clone a voice

The Mihu Voice library: voice orbs to choose from, the English female voice Emily selected, scenario chips such as order status and appointment, and her greeting ready to play.

Two models, one API

## The ears and the voice of every Mihu agent, now open to your product.

mihu-umay-v1 for speech-to-text and mihu-ses-v1 for text-to-speech sit behind one API key, over REST and WebSocket. Both answer the same way in every region and work with the official OpenAI SDKs by changing the base URL.

Umay

Speech-to-text, live and files

Ses

Text-to-speech, streamed

Languages

TR, EN, DE, FR, IT, RU, ES

Where it runs

Europe, Türkiye or on-prem

Umay · speech-to-text

## Text appears while the caller is still talking.

Partial text streams in as the person speaks. The final sentence is ready about 80 ms after they stop, quicker than a blink, so your agent can reply without a pause.

On Turkish read speech Umay has a 3.42% word error rate, the lowest of nine systems we tested. In Turkish-only mode one server handles up to 256 live calls.

Live calls over WebSocket: 16 kHz PCM in 100 ms frames, one socket

Files in any common format over REST, 19.5× faster than real time in fast mode

8 kHz phone audio supported

PARTIAL “Hi, I’d like to check my ord…” en · 0.44 s after speech starts

FINAL “Hi, I’d like to check my order status.” en · 79 ms after end of speech

S2 · FEMALE “It will be at your door tomorrow morning.” diarized · confidence 96%

FINAL “Can you send me the tracking link?” en · final text

Live call 16 kHz PCM, 100 ms frames, one socket. FINAL ~80 MS

WebSocket · live stream

REST · audio files

8 kHz phone audio

Partial and final text on a live call, with speakers tagged as they talk.

### Everything Umay can do

Speaker options are switched on per request; off means no extra latency.

File transcription Audio file to text. The most accurate choice for recordings. model=umay

Live transcription Text while the person speaks. For voice agents and calls. model=umay-stream

Asynchronous jobs The result arrives on your webhook. For long recordings. webhook_url

Number and text formatting Spoken words to written form. On by default. normalize

Hints Favours the customer names and local places you give. hints, hint_context

Language model Extra language knowledge on files of clean read speech. mode=normal

Gender detection Gender of each speaker, with confidence. For routing, analytics and addressing. gender_detection

Diarization Who said what, when two or more people speak. diarize

Speaker lock Only one person becomes text, even when someone talks nearby. speaker_lock

Voice isolation Removes other voices before recognition. voice_isolation

Reference voiceprints Follows a known person from the first second, even if someone else speaks first. reference_voiceprint

Speaker verification Is this the stored person? A helper check during a call. POST /v1/voiceprints/verify

Audio isolation Returns the recording without other voices and noise. For archives and sharing. POST /v1/audio/isolation

Aa Smart formatting “one thousand two hundred fifty dollars” → $1,250 ITN

Aa Dates and times “March third at four thirty” → 3 March, 16:30 ITN

⌕ Place-name boost Built-in lists + your own hotwords HOTWORDS

Smart formatting and place names

## Amounts, dates and addresses come out the way a CRM expects.

Amounts, dates and plates are written as numbers, adding about 6 ms. Built-in district and neighbourhood lists plus your own hotwords lift live place-name accuracy by 3 to 8 points.

1.68% word error rate on 500 spoken addresses, 8 kHz, with place hints

97.2% of address sentences fully correct with place hints

Ses · text-to-speech

## Speech that sounds local.

Ses reads addresses, place names, amounts, dates, plates and phone numbers the way people say them, in all seven languages. Audio streams back before the whole sentence is synthesised, and starts in about 0.1 s.

22 built-in voices in 7 languages, plus your own

24 kHz audio as pcm, wav, mp3, opus, aac or flac

Barge-in: closing the connection stops speech the moment the caller talks

Optional Turkish generate-and-verify, and Russian stress

Smart reading: ‘17 Oak St, Apt 4’ read as seventeen Oak Street, apartment four in English, ‘16:30’ as las cuatro y media in Spanish, and a Turkish door number, with 24 kHz streaming audio out. One request, read correctly in each language, streamed as it is generated.

Voice cloning

## Your own voice from 10–30 seconds of speech.

Record a short sample and Ses builds a custom voice you can use next to the 22 built-in ones. Every clone goes through a reading check and needs the speaker’s consent. The recording is used to build the voice and then discarded; only the voice itself is kept.

Clone a voice in the console (https://voice.mihu.ai/console)

Benchmarks · November 2026

## Accuracy in all 7 languages.

Every system is scored the same way on public test sets, with numbers and formats normalised. Lower word error rate is better.

Speech-to-text · Umay · FLEURS, clean audio

Language

Word error rate

Word accuracy

Türkçe

3.42%

96.58%

English

6.44%

93.56%

Deutsch

6.46%

93.54%

Français

6.70%

93.30%

Español

3.68%

96.32%

Italiano

3.34%

96.66%

Русский

5.61%

94.39%

Text-to-speech · Ses · read back by Whisper large-v3 at 8 kHz

Language

Word error rate

Word accuracy

Türkçe

3.38%

96.62%

English

1.29%

98.71%

Deutsch

3.62%

96.38%

Français

4.41%

95.59%

Español

2.53%

97.47%

Italiano

3.36%

96.64%

Русский

5.04%

94.96%

Ses · time to the first audio, on CPU

TR 0.10 s

EN 0.11 s

DE 0.11 s

FR 0.15 s

ES 0.12 s

IT 0.13 s

RU 0.31 s

Turkish is also measured on phone calls and broadcast media. Methods, every system tested and the full charts are on the benchmarks page. All benchmarks → (https://voice.mihu.ai/benchmarks)

The console

## Everything the API does, in the browser.

Try voices, transcribe, clone and manage keys without writing code. And you decide how long anything stays.

The Mihu Voice console Voices page: shortcuts to clone a voice, text to speech, transcribe a call and API keys, and a searchable list of built-in voices such as Alp, Kaya, Ece and Aysu with language and gender. Voices: the 22 built-in voices and your own, with shortcuts to clone, speak, transcribe and get keys.The Mihu Voice console Text to Speech page: a Turkish sentence for the voice Alp, character count and price, scenario chips, and settings for speed, natural reading of numbers and dates, pronunciation check, pronunciation dictionary and download format. Text to Speech: type, pick a voice and speed, check pronunciation and download.

### Voices

The built-in voices and your own. Listen to any of them before you use it.

### Text to Speech

Type, pick a voice and hear it start in about 0.1 s. Download as mp3, wav, flac and more.

### Speech to Text

Upload a file or a link, record, or go live and watch the words appear as you speak.

### Voice Isolator

Remove background voices and noise from a recording; compare before and after.

### Clone Voice

A voice from 10–30 seconds of speech, with a reading check and the speaker’s consent.

### API keys and usage

Keys for your servers, quick-start code per region, and what you have spent.

Integration

## Plugs into the contact centre you already run.

No Mihu SDK is required and nothing runs on your side. Live calls stream audio over a WebSocket; files and synthesis use plain HTTPS. Choosing a region changes only the host.

Europe: stt.eu.mihu.ai and tts.eu.mihu.ai

Türkiye: stt.tr.mihu.ai and tts.tr.mihu.ai, data stays in the country (KVKK)

On your own servers: Docker on CPU, offline, no GPU and no internet needed

Official OpenAI SDKs work by changing the base URL

WS /v1/stream · python import asyncio, json, websockets URL = "wss://stt.eu.mihu.ai/v1/stream?model=umay-stream&language=tr&api_key=KEY" async def transcribe(chunks): # 100 ms of 16 kHz mono int16 PCM async with websockets.connect(URL, max_size=None) as ws: async def read(): async for m in ws: d = json.loads(m) if d["type"] == "final": print(d["text"]) # send this to your LLM reader = asyncio.create_task(read()) for chunk in chunks: await ws.send(chunk) await ws.send(json.dumps({"event": "end"})) await reader

### Data protection: KVKK and GDPR

### Audio is not stored

Live streams are processed in memory and never written to disk. A file is held only while it is transcribed.

### Transcripts are not logged

API results are returned to you, not kept. In the console you choose how long transcripts stay, from not at all to 3 years.

### Voiceprints by consent

Voice recordings are never stored. Only a numeric signature is kept, and only if you ask for it; it can be deleted for good.

Pricing

## Pay for the audio you send and the characters you speak.

List prices in euros. Options are billed separately, only when you use them. Files and live streams cost the same.

Speech-to-text · mihu-umay-v1

€0.09 per hour of audio

€0.0015 per minute of audio

Text-to-speech · mihu-ses-v1

€10 per million characters

€0.01 per thousand characters

Speech-to-text options

€0.01 each

Per conversation: voice isolation, diarization, speaker lock, gender detection. Per request: voiceprints, speaker verification.

What typical workloads cost at list price

Workload

Volume

Cost

Transcribe 10,000 calls of 3 minutes

500 hours of audio

€45.00

Speak 10,000 replies of 200 characters

2 million characters

€20.00

Transcribe 1,000 hours of recordings

1,000 hours of audio

€90.00

10,000 calls of 3 minutes with speaker diarization

500 hours + 10,000 conversations

€145.00

Running on-prem or at higher volume? Talk to sales → (https://mihu.ai/contact)

## Give your product a voice and ears.

Listen to the voices, transcribe a call in the console, or get an API key and stream your first audio in minutes.

Get an API key (https://voice.mihu.ai/console/keys) Read the docs (https://developers.mihu.ai/stt/introduction)
