Research

Research backing Praxy's voice product — Indic TTS and STT for languages historically under-served by foundation models. The work below tackles two questions: how to bring a non-Indic-native TTS base to commercial-class quality on Telugu, Tamil, and Hindi without training a new acoustic decoder, and how to evaluate accent quality with metrics that don't hide behind aggregate WER.

Papers

Praxy Voice: Voice-Prompt Recovery + BUPS for Commercial-Class Indic TTS from a Frozen Non-Indic Base at Zero Commercial-Training-Data Cost

Venkata Pushpak Teja Menta · arXiv preprint · cs.SD / cs.CL / eess.AS · 2026

Commercial TTS systems produce near-native Indic audio, but the best open-source bases (Chatterbox, Indic Parler-TTS, IndicF5) trail them on measured phonological dimensions, and the most widely adopted multilingual base (Chatterbox, 23 languages) does not even tokenise Telugu or Tamil. We ask: what is the minimum intervention that brings such a non-Indic-native base to commercial-class output on Telugu, Tamil, and Hindi, without training a new acoustic decoder and without any commercial TTS training data? We combine three pieces: (1) BUPS, a Brahmic Unified Phoneme Space that deterministically romanises seven Indic scripts to ISO-15919 so Chatterbox's Latin tokeniser can process them; (2) a LoRA adapter on only the text-token predictor (Chatterbox's t3), trained on ~1,220h of licensed Indic audio with a Hindi-proxy language_id; (3) a voice-prompt recovery recipe — an 8–11s same-language reference clip plus three sampling overrides (exaggeration 0.7, temperature 0.6, min_p 0.1; "Config B") — that recovers commercial-class acoustic output with no acoustic-decoder training. On Hindi, the LoRA regresses accuracy and we instead use vanilla Chatterbox + Config B, giving a two-branch deployment. Evaluated on 10-utterance pilot sets with the companion PSP benchmark, Praxy Voice matches or slightly leads commercial baselines: 26.7% retroflex collapse on Telugu (vs Sarvam Bulbul 33.3%), 71% Tamil-zha collapse (vs commercial trio's 86%), 0.025 LLM-WER on Hindi (tied with Cartesia Sonic-3). For intra-sentential code-mix we add a third branch (IndicF5 + native-script transliteration) that drops code-mix LLM-WER from 0.80–0.85 to 0.14–0.27 across Hi/Te/Ta. We release R6 LoRA weights (Apache-2.0), inference code and router (MIT), and a Gradio demo.

arXivPDF

PSP: An Interpretable Per-Dimension Accent Benchmark for Indic Text-to-Speech

Venkata Pushpak Teja Menta · arXiv preprint · cs.SD / cs.CL · 2026

Standard TTS evaluation measures intelligibility and overall naturalness but does not quantify accent. We introduce PSP (Phoneme Substitution Profile), a benchmark that decomposes accent into six interpretable dimensions: retroflex collapse rate, aspiration fidelity, vowel-length fidelity, Tamil-zha fidelity, Frechet Audio Distance, and prosodic signature divergence. We benchmark four commercial and open-source systems across Hindi, Telugu, and Tamil, and find that retroflex collapse grows with phonological difficulty and that commercial leaders in word error rate do not uniformly lead on other accent dimensions — exposing axes of TTS quality that aggregate metrics hide.

arXivPDF

On HuggingFace

Praxy Voice — Gradio demo

Space

Live demo of Praxy Voice across Hindi, Telugu, and Tamil. Try the BUPS + LoRA + Config B pipeline end-to-end.

Praxy Voice R6

Model

LoRA adapter on Chatterbox's t3 text-token predictor, trained on ~1,220h of licensed Indic audio. Released under Apache-2.0.

PSP Native Centroids

Dataset

Native-speaker centroids used by the PSP benchmark to score accent fidelity across Hindi, Telugu, and Tamil.

Code

praxelhq/praxy

Open-source Indic TTS — Hindi, Telugu, Tamil, English. Includes BUPS romanisation, the LoRA + Config B inference router, and code-mix branch. MIT.

praxelhq/psp-eval

Reference implementation of the PSP (Phoneme Substitution Profile) benchmark — six interpretable accent dimensions across Hindi, Telugu, Tamil.