Natural, expressive speech across Indian languages
731981.4K63K134
Tags
B2C<500KProduct launchGlobalDemoFounder-led
TLVC Rating
3.80
Hook
Editing / Creativity
Copy
Sentiment of launch
Distribution strategy
Community Rating
No ratings yet
to rate this launch.
Comments (13)
Sign in to join the discussion.
R
Rhea K.20h ago
Finally a TTS that won't butcher my maasi's name on hold music. The bar was underground but still, progress.
T
Tolu Adeyemi20h ago
The waveform b-roll in the launch video is doing a LOT of heavy lifting. Give me one unedited sample reading a Tamil news bulletin and I'll believe you.
S
Siddharth V.19h ago
Pre-recorded demo or live? Give me a Colab I can hit with my own text in Marathi, Gujarati, and Assamese before I touch the API.
M
Matteo P.19h ago
Shipped something adjacent at a big co in 2020 and it still can't say 'Bengaluru' without stress-testing my blood pressure. Curious how Flash handles code-mixed Hinglish mid-sentence.
A
Akanksha R.19h ago
Indian-language voice is a category where the winner takes the stack: IVR, edtech, dubbing, agents. Been watching this team closely.
P
Pavan I.19h ago
Every Indic TTS launch sounds great in the demo reel and then whispers 'ghanta' when you feed it a real customer-support script. I'll wait for third-party samples.
N
nils19h ago
Tagline rewrite, no charge: 'Twelve languages. One voice that actually sounds like it grew up here.' Yours is fine, mine slaps.
M
Mehreen Q.19h ago
The thumbnail has the model name in three different weights and I counted two kerning crimes before the 4 second mark. Audio research team ate though.
P
Pragya S.19h ago
Are weights open? Any HF checkpoint? Pricing page is lovely but the docs link goes to a login wall and that's a vibe killer.
B
Burhan M.19h ago
'Frontier-quality speech into a smaller model' is the AI equivalent of 'spacious one-bedroom in Bandra.' Prove it with latency numbers on a Jio connection.
W
Wenhao Liang19h ago
What's the vocoder situation here, and did you train on prosody-annotated data or let the model figure out intonation from raw audio? Asking because Indic prosody breaks most Western-trained stacks.
I
Ira M.19h ago
I've been 3 weeks away from launching my regional-language audiobook thing for 9 months. Guess I'm about to be 3 weeks away from integrating Bulbul instead.
J
jaskiran19h ago
Building in the dubbing space and genuinely rooting for you all. The expressiveness gap on Indian languages has been embarrassing for years, nice to see someone actually attack it.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Sarvam
Natural, expressive speech across Indian languages
Natural, expressive speech across Indian languages
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Comments (13)
Finally a TTS that won't butcher my maasi's name on hold music. The bar was underground but still, progress.
The waveform b-roll in the launch video is doing a LOT of heavy lifting. Give me one unedited sample reading a Tamil news bulletin and I'll believe you.
Pre-recorded demo or live? Give me a Colab I can hit with my own text in Marathi, Gujarati, and Assamese before I touch the API.
Shipped something adjacent at a big co in 2020 and it still can't say 'Bengaluru' without stress-testing my blood pressure. Curious how Flash handles code-mixed Hinglish mid-sentence.
Indian-language voice is a category where the winner takes the stack: IVR, edtech, dubbing, agents. Been watching this team closely.
Every Indic TTS launch sounds great in the demo reel and then whispers 'ghanta' when you feed it a real customer-support script. I'll wait for third-party samples.
Tagline rewrite, no charge: 'Twelve languages. One voice that actually sounds like it grew up here.' Yours is fine, mine slaps.
The thumbnail has the model name in three different weights and I counted two kerning crimes before the 4 second mark. Audio research team ate though.
Are weights open? Any HF checkpoint? Pricing page is lovely but the docs link goes to a login wall and that's a vibe killer.
'Frontier-quality speech into a smaller model' is the AI equivalent of 'spacious one-bedroom in Bandra.' Prove it with latency numbers on a Jio connection.
What's the vocoder situation here, and did you train on prosody-annotated data or let the model figure out intonation from raw audio? Asking because Indic prosody breaks most Western-trained stacks.
I've been 3 weeks away from launching my regional-language audiobook thing for 9 months. Guess I'm about to be 3 weeks away from integrating Bulbul instead.
Building in the dubbing space and genuinely rooting for you all. The expressiveness gap on Indian languages has been embarrassing for years, nice to see someone actually attack it.