Signal Processing, Interpretation and REpresentation (SPIRE) Laboratory, Indian Institute of Science, Bangalore.
Bangalore, India
Joined January 2020
- Tweets166
- Following13
- Followers110
- Likes54
๐๏ธ Turn Your Voice into an Opportunity!
๐ฒ Scan the QR code on the poster and apply today!
โจ Your Voice Can Create Real Change!
#WeAreHiring #SPIRELab #IISc #PartTimeJobs #SpeechRecording #ArtificialIntelligence #AIResearch #WorkFromAnywhere
This Durga Puja, Project Vaani celebrates India's linguistic diversity and our collective journey towards a future where AI can listen to and understand more of India's voices.
Every celebration has a voice. โค๏ธ
From Bengali and Assamese to Odia, Kokborok and Chakma, every language carries a unique story, culture and identity.
Wishing everyone a joyful and blessed Durga Puja! โจ
#ProjectVaani #LinguisticDiversity #InclusiveAI #SpireLab #ARTPARK
๐๐๐ถ๐น๐ฑ๐ถ๐ป๐ด ๐ป๐ผ๐ถ๐๐ฒ-๐ฟ๐ผ๐ฏ๐๐๐ ๐๐ฝ๐ฒ๐ฒ๐ฐ๐ต ๐๐? ๐๐ผ๐๐ป๐น๐ผ๐ฎ๐ฑ ๐๐ต๐ฒ ๐ฉ๐ฎ๐ฎ๐ป๐ถ ๐ก๐ผ๐ถ๐๐ฒ ๐๐๐ฒ๐ป๐ ๐๐ฎ๐๐ฎ๐๐ฒ๐ ๐ฎ๐ป๐ฑ ๐๐๐ฎ๐ฟ๐ ๐ฒ๐
๐ฝ๐ฒ๐ฟ๐ถ๐บ๐ฒ๐ป๐๐ถ๐ป๐ด. Follow the Datathon as Phase 2 heads to the finish. Links in the comments.
Thank you to the IndoML 2026 organisers, the Datathon chairs, and every team taking on the challenge.
๐ง๐ต๐ฒ ๐ฑ๐ฎ๐๐ฎ ๐ฏ๐ฒ๐ต๐ถ๐ป๐ฑ ๐ถ๐:
The challenge runs on the Vaani Noise Event Dataset, our open-source collection of real Indian speech recorded with real background noise, with every noise event manually timestamped to the millisecond.
๐ช๐ต๐ฎ๐โ๐ ๐ป๐ฒ๐
๐:
Phase 2 closes on 17 October, and the top teams will present at IndoML 2026 in Kolkata this December.
๐ง๐ต๐ฒ ๐ฟ๐ฒ๐๐ฝ๐ผ๐ป๐๐ฒ:
250+ teams from 35+ institutions registered, from students to industry engineers.
๐ง๐ต๐ฒ ๐ฐ๐ต๐ฎ๐น๐น๐ฒ๐ป๐ด๐ฒ:
Track 1: Detect noise events in real Indian speech, with precise start and end times.
Track 2: Use those timestamps to remove the noise while keeping the speech intact.
Now in its 7th edition, IndoML brings together leading institutions and research labs from India and abroad.
๐ฆ๐ถ๐ป๐ฐ๐ฒ ๐ฎ๐ฌ๐ฎ๐ฌ, ๐๐ป๐ฑ๐ผ๐ ๐ ๐ต๐ฎ๐ ๐ฏ๐ฟ๐ผ๐๐ด๐ต๐ ๐๐ผ๐ฟ๐น๐ฑ-๐ฐ๐น๐ฎ๐๐ ๐ ๐ ๐ฟ๐ฒ๐๐ฒ๐ฎ๐ฟ๐ฐ๐ต ๐๐ผ ๐๐ป๐ฑ๐ถ๐ฎ, ๐บ๐ฒ๐ป๐๐ผ๐ฟ๐ฒ๐ฑ ๐ถ๐๐ ๐ป๐ฒ๐
๐ ๐ด๐ฒ๐ป๐ฒ๐ฟ๐ฎ๐๐ถ๐ผ๐ป ๐ผ๐ณ ๐ฟ๐ฒ๐๐ฒ๐ฎ๐ฟ๐ฐ๐ต๐ฒ๐ฟ๐, ๐ฎ๐ป๐ฑ ๐ฐ๐ผ๐ป๐ป๐ฒ๐ฐ๐๐ฒ๐ฑ ๐ฎ๐ฐ๐ฎ๐ฑ๐ฒ๐บ๐ถ๐ฎ ๐๐ถ๐๐ต ๐ถ๐ป๐ฑ๐๐๐๐ฟ๐.
What happens to a voice assistant when a pressure cooker whistles mid-sentence?
For most speech models, it falls apart. Thatโs the problem the IndoML team set for 250+ teams at the IndoML 2026 Datathon.
#IndoML2026 #SpeechAI #IndicLanguages #OpenData #ResponsibleAI
Come say hello, grab some exciting goodies, and find out how you can be part of this national initiative.
#Interspeech2026 #Vaani #SpeechAI #IndicLanguages
๐๐ก๐๐ซ๐ ๐ฒ๐จ๐ฎ'๐ฅ๐ฅ ๐๐ข๐ง๐ ๐ฎ๐ฌ:
๐ ๐๐ญ๐๐ง๐ ๐4, ๐ฐ๐ข๐ญ๐ก ๐๐จ๐จ๐ ๐ฅ๐
We'll be at the booth all week, showing Vaani's open speech datasets, models and demos built for India's languages.
๐น ๐๐ก๐๐ญ ๐๐ซ๐๐๐ญ๐ก ๐๐จ๐ฎ๐ง๐๐ฌ ๐๐๐ฏ๐๐๐ฅ ๐๐๐จ๐ฎ๐ญ ๐๐ข๐ซ๐๐ฅ๐จ๐ฐ ๐๐ข๐ฆ๐ข๐ญ๐๐ญ๐ข๐จ๐ง ๐ข๐ง ๐๐ฌ๐ญ๐ก๐ฆ๐
Lung sound patterns track spirometric airflow limitation in asthma. How strongly they track it depends on chest location, frequency band, age and gender.
1. ๐๐๐ง๐๐๐ซ ๐๐ข๐๐ฌ ๐ข๐ง ๐๐๐ ๐๐ฎ๐ง๐ฌ ๐๐๐๐ฉ๐๐ซ ๐๐ก๐๐ง ๐ญ๐ก๐ ๐๐๐ญ๐
Across training paradigms, balancing fine-tuning data doesn't remove gender bias from pretrained speech recognition. The real fix lies in representations.
Interspeech 2026 is here, and SPIRE Lab will be at ICC Sydney from 27 Sep to 1 Oct.
๐๐ก๐๐ญ ๐ฐ๐'๐ฅ๐ฅ ๐๐ ๐ฉ๐ซ๐๐ฌ๐๐ง๐ญ๐ข๐ง๐ :
๐ Read the full conversation in Moneycontrol: moneycontrol.com/artificial-โฆ
#ProjectVaani #Sravaani #LanguageAI #IndianLanguages #AIforIndia #SpeechAI #IISc #ARTPARK #SPIRELAB
Truly inclusive AI takes more than good intentions. It takes sustained investment, collaboration, and the representation of every voice.