Pal · ConvFill Talker + Gemma 4 Reasoner

हिन्दी · मराठी · English · ગુજરાતી · தமிழ் · తెలుగు · മലയാളം · ಕನ್ನಡ — on Hugging Face's speech-to-speech

Models ConvFill Talker (multilingual) Gemma 4 (Reasoner) Pocket-TTS multilingual Pocket-TTS Gujarati faster-whisper (STT) Built by Hugging Face andito tfrere A-Mahla Indic build DocMode Health Original by University of Washington UbiComp Lab

Tap to start

Conversation

About

Pal speaks Hindi, Marathi, English, Gujarati, Tamil, Telugu, Malayalam and Kannada in one cloned voice. A Gemma 4 Reasoner streams knowledge to a multilingual ConvFill Talker, so Pal starts answering immediately while the facts arrive. Noise suppression, echo rejection and a backchannel-aware interruption gate keep Pal from mistaking its own voice, "haan" or background noise for a new question.

View the project on GitHub

The pipeline

  1. You speak
  2. VAD denoises (RNNoise), detects speech, ignores echo silero-vad
  3. STT transcribes it faster-whisper small · 8 Indian languages
  4. Reasoner streams concise knowledge Gemma 4 (OpenAI-compatible endpoint)
  5. Talker turns knowledge into the spoken reply paulsonpaul/Qwen3-0.6B-ConvFill-multi
  6. TTS speaks it back in Pal's voice Pocket-TTS multilingual + Gujarati (CPU)
  7. The orb replies

Settings

Noise gate Off
Off −3 dB
Mutes the mic below the handle so room noise isn't sent. Slide fully left to turn it off; the bar shows your live input while in a call.
Connect first, then come back here to apply live changes.

Tools

Let the assistant act during the conversation. Changes apply live.

Web search Look things up on Google, via Serper.
Camera Let the assistant see through your webcam.
A live preview shows bottom-left while the camera is on.

That's a wrap for now