Hinglish and beyond
Real speech mixes languages mid-sentence. Pāṇini detects the switch inline and keeps one continuous voice across scripts, so Hinglish, Tanglish or Arabic-French code mixing never breaks the delivery.
Developers · Pāṇini text to speech
One speech model that never switches accents on you.
Code switching, emotion, non-verbal expression and custom pronunciation, in 250+ languages from a single endpoint.
250+
Languages, one model
120
Platform voices
< 200ms
Time to first byte
Try it
Pick a language, pick a voice, hear the difference.
39 chars39/220
Pāṇini
Voices
Capabilities
Everything speech needs to sound like a person.
Each control is a field on the same request body. No separate models, no per-feature endpoints.
250+ languages, one voice
Set language to auto or pin it per request. The same voice identity carries across every language without re-cloning or a separate model per locale.
Tone that fits the moment
Steer delivery with emotion and style controls: empathetic for support, assertive for collections, cheerful for onboarding. Intensity is adjustable per request.
laugh, sigh, hm, aha
Inline tags render breaths, laughs, sighs, hesitations and gasps as part of the audio stream. No extra calls, no stitched clips.
Read the way people say it
IFSC, KYC, EMI, GST, SKU and dates, currencies and phone numbers are normalised per language before synthesis, so nothing gets spelled out awkwardly.
Your names, said correctly
Upload a pronunciation dictionary for brand names, product SKUs, drug names or local place names. Entries apply across every language and every voice.
Fine control per request
Adjust speed, pitch and pause length without artefacts. Useful for IVR prompts, audiobooks and telephony where pacing matters.
Sentence by sentence
Audio streams as it generates, so playback begins before the full response is synthesised. WAV, MP3, PCM and μ-law for telephony.
POST /v1/speech/stream
Audio is generated sentence by sentence and streamed back as a chunked response, so the first bytes arrive long before the last word is synthesised. Use this for agents, telephony and anything a person is waiting on.
Keys start with vsk_ and are sent as a bearer token. Every voice speaks every language, so the only required fields are text and voice_id.
stream.sh
1curl -N -X POST https://api.sonexlabs.com/v1/speech/stream \2 -H "Authorization: Bearer $SONEX_API_KEY" \3 -H "Content-Type: application/json" \4 -d '{5 "text": "नमस्ते! आपका order कल deliver हो जाएगा.",6 "voice_id": "72ly9crx9v",7 "output_format": "wav"8 }' \9 --output speech.wavVoices
120 platform voices, plug and play.
Every voice speaks all 250+ languages while keeping its own timbre. Need something specific? Clone a voice from ten seconds of audio.
Priya
FemaleIndian · Warm, conversational
Arjun
MaleIndian · Calm, assured
Meera
FemaleIndian · Bright, upbeat
Kabir
MaleIndian · Deep, narrative
Ananya
FemaleIndian · Crisp, professional
Olivia
FemaleAmerican · Neutral, clear
James
MaleBritish · Measured, editorial
Sofia
FemaleLatin American · Friendly, natural
Yuki
FemaleJapanese · Soft, polite
Amara
FemaleWest African · Rich, grounded
Omar
MaleLevantine · Steady, formal
Nova
NeutralGlobal · Even, brandless
+ 108 more platform voices
Language coverage
Every language on this list costs the same.
Twenty four scripts below, and the full register of 250 languages underneath. Most of them have never had a production voice model before.
भोजपुरी
Bhojpuri
मारवाड़ी
Marwari
गढ़वळि
Garhwali
سرائیکی
Saraiki
हिन्दी
Hindi
தமிழ்
Tamil
తెలుగు
Telugu
বাংলা
Bengali
मराठी
Marathi
ગુજરાતી
Gujarati
ਪੰਜਾਬੀ
Punjabi
ಕನ್ನಡ
Kannada
മലയാളം
Malayalam
ଓଡ଼ିଆ
Odia
অসমীয়া
Assamese
संस्कृतम्
Sanskrit
اردو
Urdu
नेपाली
Nepali
Yorùbá
Yoruba
Kiswahili
Swahili
isiZulu
Zulu
Soomaali
Somali
Dagbani
Dagbani
العربية
Arabic
中文
Chinese
The register
220 of 250 languages listed
A
- Abkhazian
- Afrikaans
- Albanian
- Algerian Arabic
- Amdo Tibetan
- Amharic
- Angika
- Armenian
- Assamese
- Asturian
- Atayal
- Ayacucho Quechua
- Azerbaijani
B
- Baatonum
- Bafia
- Bafut
- Bamun
- Banjar
- Baoulé
- Basa Cameroon
- Bashkir
- Basque
- Belarusian
- Bengali
- Bhili
- Bhojpuri
- Bodo
- Bosnian
- Braj
- Breton
- Buginese
- Bulgarian
- Bundeli
- Burmese
C
- Cameroon Pidgin
- Cantonese
- Catalan
- Cebuano
- Central Nahuatl
- Central Yupik
- Chadian Arabic
- Chichewa
- Chimborazo Highland Quichua
- Chinese
- Chuvash
- Cornish
- Croatian
- Czech
D
- Dagbani
- Danish
- Dogri
- Dotyali
- Dutch
- Dyula
E
- Eastern Balochi
- Eastern Mari
- Egyptian Arabic
- Embu
- English
- Erzya
- Esperanto
- Estonian
- Extremaduran
F
- Farefare
- Filipino
- Finnish
- French
- Fulah
G
- Galician
- Ganda
- Garhwali
- Georgian
- German
- Goan Konkani
- Greek
- Guarani
- Gujarati
- Gujari
- Gulf Arabic
- Gusii
H
- Hadothi
- Haitian
- Hausa
- Hawaiian
- Hebrew
- Hindi
- Huautla Mazatec
- Hungarian
I
- Icelandic
- Igbo
- Indonesian
- Interlingua
- Inupiaq
- Irish
- Iron Ossetic
- Italian
J
- Japanese
- Javanese
K
- Kabyle
- Kalenjin
- Kamba
- Kannada
- Kashmiri
- Kazakh
- Khams Tibetan
- Khmer
- Kinnauri
- Kinyarwanda
- Kirghiz
- Konkani
- Korean
- Kuanyama
L
- Lao
- Latgalian
- Latvian
- Levantine Arabic
- Libyan Arabic
- Ligurian
- Lingala
- Lithuanian
- Luba-Lulua
- Luo
- Lushai
- Luxembourgish
M
- Macedo-Romanian
- Macedonian
- Maithili
- Malay
- Malayalam
- Manipuri
- Manx
- Maori
- Marathi
- Marwari
- Meru
- Mesopotamian Arabic
- Mewari
- Min Nan Chinese
- Moksha
- Mongolian
- Moroccan Arabic
N
- Najdi Arabic
- Ndonga
- Neapolitan
- Nepali
- Nigerian Pidgin
- Nimadi
- Norwegian
O
- Occitan
- Odia
- Omani Arabic
- Oromo
P
- Pahari-Potwari
- Paiwan
- Panjabi
- Persian
- Piemontese
- Plateau Malagasy
- Polish
- Portuguese
R
- Rangi
- Romanian
- Romansh
- Rombo
- Russian
S
- Sakizaya
- Sanskrit
- Santali
- Saraiki
- Sardinian
- Sediq
- Serbian
- Shona
- Sicilian
- Sikkimese
- Sindhi
- Sinhala
- Slovak
- Slovenian
- Somali
- Soninke
- Spanish
- Standard Arabic
- Sudanese Arabic
- Swahili
- Swedish
T
- Taita
- Tajik
- Tamil
- Tatar
- Telugu
- Thai
- Tibetan
- Tigrinya
- Tlingit
- Toki Pona
- Tswana
- Tulu
- Tunisian Arabic
- Turkish
- Turkmen
- Twi
U
- Udmurt
- Uighur
- Ukrainian
- Umbundu
- Upper Sorbian
- Urdu
- Uzbek
V
- Vietnamese
- Võro
W
- Welsh
- Western Mari
- Wolof
X
- Xhosa
Y
- Yakut
- Yangben
- Yaqui
- Yoruba
Z
- Zarma
Why one model matters
The world speaks 7,000 languages. Voice AI speaks about thirty.
Most multilingual TTS is English first, with other languages bolted on afterwards. Pāṇini was multilingual from the first training run, so latency, price and quality do not change when the language does.
~30
Languages typical multilingual voice AI covers
~4B
People whose first language falls outside that ~30
1,300+
Languages spoken across India alone
250+
Languages Pāṇini covers at one flat rate
Scale
Built for production, not for demos.
< 200 ms
Time to first byte
250+
Languages, one endpoint
120
Platform voices, plug and play
36+
Indian languages and dialects
8 kHz
Telephony-grade training
10,000
Concurrent streams
Pricing
Per character, one rate.
Bhojpuri costs what English costs. No language tiers, no minimum commitment.
Pāṇini TTS API$0.0313 / min
1,000,000 characters, roughly 20 hours of audio~$52
Free to start10,000 chars
Ship a voice your users recognise as their own.
Free tier, no card, 10,000 characters included. Every language available from the first request.