Developers · Pāṇini text to speech

Text to speech API for 250+ languages

Pāṇini TTS turns text into speech from one endpoint, including mixed-language text such as Hinglish. Paste the Python on the right, add your key and run it.

Free API tier: 10,000 free text-to-speech characters and one free voice clone that speaks every language, no card required.

Then: $0.03125 per 1,000 characters, the same rate for every language.

Get a free API key Read the docs

speak.py

1# pip install requests
2import requests
3
4r = requests.post(
5    "https://api.sonexlabs.com/v1/speech",
6    headers={"Authorization": "Bearer YOUR_API_KEY"},
7    json={
8        "text": "नमस्ते! आपका order कल शाम तक पहुँच जाएगा।",
9        "output_format": "mp3",      # wav, mp3, ogg, opus or mulaw
10        # "voice_id": "...",         # optional: GET /v1/voices lists them
11    },
12)
13r.raise_for_status()
14with open("hello.mp3", "wb") as f:
15    f.write(r.content)
16print("characters billed:", r.headers["X-Chars-Billed"])

Only text is required. Text is capped at 5,000 characters per request here; use the streaming endpoint below for longer scripts.

Try it

Pick a language, pick a voice, listen.

39 chars · about 3 s39/220

Pāṇini

Voices

Capabilities

What the API does today.

Each option is a field on the same request body. No separate models, no per-feature endpoints.

Mixed-language text

Hinglish and beyond

Pāṇini TTS reads mixed-language text such as Hinglish naturally, in one voice.

Any voice, any language

Language comes from the text

One voice can speak across languages, and a cloned voice carries across languages too. There is no language setting to manage: the language is read from the text itself.

Voice cloning

From a 10-second sample

Upload a short reference recording to POST /v1/voices/clone and use the new voice id in any request. The speaker's accent carries over to every language the clone reads.

Streaming

Sentence by sentence

POST /v1/speech/stream returns audio as it is generated, for text up to 20,000 characters. Pāṇini TTS starts streaming audio in under 200 ms (time to first byte).

Phone audio

8 kHz and mu-law

Pāṇini TTS is trained for 8 kHz phone audio. Request 8000 Hz and mulaw output for telephony bridges, or 16 to 48 kHz for apps and video.

Formats and speed

WAV, MP3, OGG, Opus, mu-law

Pick the output format and sample rate per request, and set speed from 0.5 to 2.0. Every response reports the characters billed in the X-Chars-Billed header.

Stream long text in Python

stream.py

1# pip install requests
2import requests
3
4with requests.post(
5    "https://api.sonexlabs.com/v1/speech/stream",
6    headers={"Authorization": "Bearer YOUR_API_KEY"},
7    json={"text": open("chapter.txt", encoding="utf-8").read(), "sample_rate": 24000},
8    stream=True,
9) as r:
10    r.raise_for_status()
11    with open("chapter.wav", "wb") as f:
12        for chunk in r.iter_content(chunk_size=None):   # audio arrives sentence by sentence
13            f.write(chunk)

Phone-ready audio with cURL

prompt.sh

1curl https://api.sonexlabs.com/v1/speech \
2  --header 'Authorization: Bearer YOUR_API_KEY' \
3  --header 'Content-Type: application/json' \
4  --data '{"text": "Your appointment is at 10 am tomorrow.", "output_format": "mulaw", "sample_rate": 8000}' \
5  --output prompt.wav

Voices

Dozens of platform voices, or your own.

A voice is a recording of one speaker, and the language comes from your text, so any voice can read any supported language. The accent comes from the speaker. Pāṇini TTS has voices recorded by native speakers for Hindi, Tamil, Telugu, Bengali, Marathi, Japanese, English, French, Spanish (Latin American) and Modern Standard Arabic. For other languages, the text is read correctly by a voice from another language; for a local accent, clone a native speaker from a 10-second sample.

List the voices available to your workspace with GET /v1/voices and pass the id as voice_id.

Languages

Text to speech by language.

Every language costs the same. Each page has a live playground and notes on how that language is written and read.

See every language grouped by region on text to speech by language, or the full register below.

The register

223 of 250 languages listed

Abkhazian
Afrikaans
Albanian
Algerian Arabic
Amdo Tibetan
Amharic
Angika
Armenian
Assamese
Asturian
Atayal
Ayacucho Quechua
Azerbaijani

Baatonum
Bafia
Bafut
Bamun
Banjar
Baoulé
Basa Cameroon
Bashkir
Basque
Belarusian
Bengali
Bhili
Bhojpuri
Bodo
Bosnian
Braj
Breton
Buginese
Bulgarian
Bundeli
Burmese

Cameroon Pidgin
Cantonese
Catalan
Cebuano
Central Nahuatl
Central Yupik
Chadian Arabic
Chhattisgarhi
Chichewa
Chimborazo Highland Quichua
Chinese
Chuvash
Cornish
Croatian
Czech

Dagbani
Danish
Dogri
Dotyali
Dutch
Dyula

Eastern Balochi
Eastern Mari
Egyptian Arabic
Embu
English
Erzya
Esperanto
Estonian
Extremaduran

Farefare
Filipino
Finnish
French
Fulah

Galician
Ganda
Garhwali
Georgian
German
Goan Konkani
Greek
Guarani
Gujarati
Gujari
Gulf Arabic
Gusii

Hadothi
Haitian
Haryanvi
Hausa
Hawaiian
Hebrew
Hindi
Huautla Mazatec
Hungarian

Icelandic
Igbo
Indonesian
Interlingua
Inupiaq
Irish
Iron Ossetic
Italian

Japanese
Javanese

Kabyle
Kalenjin
Kamba
Kannada
Kashmiri
Kazakh
Khams Tibetan
Khmer
Kinnauri
Kinyarwanda
Kirghiz
Konkani
Korean
Kuanyama

Lao
Latgalian
Latvian
Levantine Arabic
Libyan Arabic
Ligurian
Lingala
Lithuanian
Luba-Lulua
Luo
Lushai
Luxembourgish

Macedo-Romanian
Macedonian
Maithili
Malay
Malayalam
Manipuri
Manx
Maori
Marathi
Marwari
Meru
Mesopotamian Arabic
Mewari
Min Nan Chinese
Moksha
Mongolian
Moroccan Arabic

Najdi Arabic
Ndonga
Neapolitan
Nepali
Nigerian Pidgin
Nimadi
Norwegian

Occitan
Odia
Omani Arabic
Oromo

Pahari-Potwari
Paiwan
Panjabi
Persian
Piemontese
Plateau Malagasy
Polish
Portuguese

Rangi
Romanian
Romansh
Rombo
Russian

Sakizaya
Sanskrit
Santali
Saraiki
Sardinian
Sediq
Serbian
Shona
Sicilian
Sikkimese
Sindhi
Sinhala
Slovak
Slovenian
Somali
Soninke
Spanish
Standard Arabic
Sudanese Arabic
Swahili
Swedish
Sylheti

Taita
Tajik
Tamil
Tatar
Telugu
Thai
Tibetan
Tigrinya
Tlingit
Toki Pona
Tswana
Tulu
Tunisian Arabic
Turkish
Turkmen
Twi

Udmurt
Uighur
Ukrainian
Umbundu
Upper Sorbian
Urdu
Uzbek

Vietnamese
Võro

Welsh
Western Mari
Wolof

Xhosa

Yakut
Yangben
Yaqui
Yoruba

Zarma

Choosing an API

How to choose a text to speech API.

We sell one, so read this as a checklist to apply to every option, us included. The right choice depends on your languages and your channel, not on a ranking.

Listen in your languages, with your own text

A demo sentence hides the hard parts. Paste real lines with prices, dates, phone numbers, product names and mixed scripts, and listen to each API in every language you need.

Ask which languages have a native voice

Many APIs, this one included, read some languages with a voice recorded in another language, which carries that speaker's accent. Pāṇini TTS has native voices for Hindi, Tamil, Telugu, Bengali, Marathi, Japanese, English, French, Spanish (Latin American) and Modern Standard Arabic. Elsewhere, clone a native speaker.

Check mixed-language text stays in one voice

If your text mixes Hindi and English, check the voice does not switch accent or speaker halfway through a sentence.

Measure time to first audio yourself

For agents and IVR, what matters is how soon audio starts, measured from the region your servers run in, not the vendor's own benchmark.

Compare how characters are billed

Per character, per credit and per minute of audio are hard to compare. Find out whether spaces, markup and retries count, and whether every language costs the same.

Check formats and rights

Phone systems need 8 kHz or mu-law; video needs 44.1 or 48 kHz. For cloning, check what consent the vendor asks for and who may use the voice.

Pricing

Per character, one rate.

Bhojpuri costs what English costs. No language tiers, no minimum commitment.

Pāṇini TTS API$0.03125 / 1K chars

1,000,000 characters$31.25

Free to start10,000 chars

Start free Full pricing

FAQ

Text to speech API, answered.

Yes. 10,000 free text-to-speech characters and one free voice clone that speaks every language, no card required. After that, Pāṇini TTS costs $0.03125 per 1,000 characters, the same rate for every language.

There is no single best one: it depends on your languages, your channel (phone, app or video) and your budget. Use the checklist on this page and test each option on your own text. We sell Pāṇini TTS, so we are not neutral.

POST your text as JSON to /v1/speech with requests and write the response body to a file, as in the snippet at the top of this page. Add output_format for MP3, Opus or mu-law, and voice_id to pick a voice.

No. Pāṇini TTS reads the language from the text. The API accepts a language field, but it has no effect on pronunciation, so you can leave it out.

Yes. Pāṇini TTS is trained for 8 kHz phone audio. Ask for output_format mulaw and sample_rate 8000 to get G.711 mu-law audio in a WAV container.

Yes, from a 10-second sample, and the free tier includes one clone. The cloned voice can read any supported language; its accent stays the reference speaker's.

Need the other direction? Valmiki speech to text transcribes audio, billed by the second.

Make your first audio file in a minute.

Free tier: 10,000 free text-to-speech characters and one free voice clone that speaks every language, no card required.

Get a free API key Read the docs