⚡ Speech → IndicConformer → emotion2vec+ (Emotion SER) → IndicTrans2 200M → Emotion-Aware TTS → Speech ● 100% OFFLINE MULTILINGUAL AI
STAGE 1
Speech Input
🎙 Raw Speech (16kHz WAV)
Live Mic / Buffer
➔
STAGE 2 (ASR)
IndicConformer ASR
AI4Bharat IndicConformer (22 Langs)
Speech → Source Transcript
➔
STAGE 3 (SER)
emotion2vec+ Base
emotion2vec+ Base (Distilled ONNX)
Raw Waveform → Real Probabilities
➔
STAGE 4 (MT)
IndicTrans2 200M
AI4Bharat IndicTrans2 (All Indic Langs)
Source Text → Translated Text
➔
STAGE 5 (TTS)
Emotion-Aware TTS
Emotion-Aware IndicTTS (Prosody Modulated)
Target Text + Emotion → Speech
➔
STAGE 6
Speech Output
🔊 Spoken Output
Speaker / Headphone Audio
INDIC-TRANSLATOR // MK-1 (SER ENABLED)
● OFFLINE LOCAL EMOTION: NEUTRAL (85.0%) 00:00:00
Audio Ready
SOURCE TRANSCRIPT ASR: IndicConformer INT8
Press Push-to-Talk or select a benchmark audio sample below.
TRANSLATION (EMOTION PRESERVED) MT: IndicTrans2 INT8
Target translation will appear here offline.
⚡ TAP PRESET SAMPLE (1-CLICK TEST):
Audio Input Source

Click Push To Talk to record live Hindi or English speech from your microphone. When stopped, the raw audio is converted into 16kHz PCM and fed into the local pipeline.

Local Pipeline Stage Execution (8 Modular Steps) 100% On-Device
1
Step 1: Audio Captured
Input: Raw 16kHz Mono PCM Audio Stream
PCM 16kHz
2
Step 2: ASR Running
Model: AI4Bharat IndicConformer (22 Langs)
ASR MODEL
3
Step 3: Transcript Produced
Output: Recognized Source Words
TEXT OUT
4
Step 4: Emotion Analysis
Model: emotion2vec+ Base (Distilled ONNX) | Result: NEUTRAL (85.0%)
EMOTION SER
5
Step 5: Translation Running
Model: AI4Bharat IndicTrans2 (All Indic Langs)
MT MODEL
6
Step 6: Translation Produced
Output: Target Language Script
TARGET TEXT
7
Step 7: TTS Running
Model: Emotion-Aware IndicTTS (Prosody Modulated)
TTS MODEL
8
Step 8: Audio Output
Output: Spoken 22.05kHz WAV Audio
SPEAKER OUT