SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2604.25441 · 2026

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 · PDF · Open in the Atlas

Code that ran

2 of 13 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 1 failure of our instrument; 11 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2604.25441.json · badge: /api/ran/2604.25441.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
praxelhq/praxy canonical (classifier label)
classifier basis: census
2 ran · 11 no recorded run
FunctionStatusWhere it lives
is_codemix Ran
ran, violated the contract we drafted
praxelhq/praxy/serving/praxy_router.py
pointer only (licence: NOASSERTION) · get_code("274cde16976dc76c")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, praxelhq/praxy (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

transliterate_codemix Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
praxelhq/praxy/serving/codemix_to_native_script.py
pointer only (licence: NOASSERTION) · get_code("915b29dadb946782")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, praxelhq/praxy (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

cardinal_te Not yet run
unverified (harvested, no recorded run)
praxelhq/praxy/praxy/linguistics/te_numbers.py
pointer only (licence: NOASSERTION) · get_code("9e6cd1c338e48ce5")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, praxelhq/praxy (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

detect_script Not yet run
unverified (harvested, no recorded run)
praxelhq/praxy/praxy/linguistics/bups.py
pointer only (licence: NOASSERTION) · get_code("77a497ab5b45a111")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, praxelhq/praxy (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

normalize_indic_text Not yet run
unverified (harvested, no recorded run)
praxelhq/praxy/praxy/linguistics/indic_numbers.py
pointer only (licence: NOASSERTION) · get_code("9f99eeb643310aa7")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, praxelhq/praxy (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

normalize_te_digits Not yet run
unverified (harvested, no recorded run)
praxelhq/praxy/praxy/linguistics/indic_numbers.py
pointer only (licence: NOASSERTION) · get_code("9dd1a325cfaecba2")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, praxelhq/praxy (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

normalize_te_digits Not yet run
unverified (harvested, no recorded run)
praxelhq/praxy/praxy/linguistics/te_numbers.py
pointer only (licence: NOASSERTION) · get_code("e076c07d4ccec617")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, praxelhq/praxy (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

ordinal_te Not yet run
unverified (harvested, no recorded run)
praxelhq/praxy/praxy/linguistics/te_numbers.py
pointer only (licence: NOASSERTION) · get_code("e69446ac8b32ec16")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, praxelhq/praxy (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

phoneme_id Not yet run
unverified (harvested, no recorded run)
praxelhq/praxy/praxy/linguistics/bups.py
pointer only (licence: NOASSERTION) · get_code("3732e394b4da8e74")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, praxelhq/praxy (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

route Not yet run
unverified (harvested, no recorded run)
praxelhq/praxy/serving/praxy_router.py
pointer only (licence: NOASSERTION) · get_code("20ed7323c70bf96a")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, praxelhq/praxy (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

segment_by_script Not yet run
unverified (harvested, no recorded run)
praxelhq/praxy/praxy/linguistics/bups.py
pointer only (licence: NOASSERTION) · get_code("072ee9b0c4da42ef")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, praxelhq/praxy (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

synthesize Not yet run
unverified (harvested, no recorded run)
praxelhq/praxy/serving/praxy_router.py
pointer only (licence: NOASSERTION) · get_code("fae81873945159d7")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, praxelhq/praxy (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

transliterate_test_set Not yet run
unverified (harvested, no recorded run)
praxelhq/praxy/serving/codemix_to_native_script.py
pointer only (licence: NOASSERTION) · get_code("f4f74c39a36c55c9")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, praxelhq/praxy (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Also present in other papers' code

Of 13 samples lifted from this paper's linked repositories, 13 carry a function identity and 0 do not. Of the 13 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 13 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Commercial text-to-speech (TTS) systems produce near-native Indic audio today, but the best open-source bases (ResembleAI Chatterbox, Indic Parler-TTS, IndicF5) trail them by meaningful margins on measured phonological dimensions -and the most widely adopted multilingual open-source base (Chatterbox, 23 languages) does not even tokenise Telugu or Tamil at inference. We ask: what is the minimum intervention that brings such a non-Indic-native multilingual base to commercial-class output on three tier-1 Indic languages (Telugu, Tamil, Hindi), without training a new acoustic decoder and without any commercial TTS training data? We answer with three engineering pieces combined into a single recipe: (1) BUPS, a Brahmic Unified Phoneme Space that deterministically romanises Devanagari, Telugu, Tamil, Kannada, Bengali, Gujarati, and Malayalam 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 via BUPS-preprocessed Indic text with a Hindi-proxy language_id on ∼1,220 h of licensed Indic audio (IndicTTS, Rasa, FLEURS, Shrutilipi); (3) an inference-time voice-prompt recovery recipe that supplies a same-language reference voice clip of 8-11 s plus three sampling overrides (exaggeration 0.7, temperature 0.6, min_p 0.1; we call these Config B) -recovering commercial-class acoustic output without any acoustic-decoder training. We additionally show that on Hindi, which Chatterbox natively covers, our LoRA adapter regresses semantic accuracy, and the same Config B + voice-prompt recipe applied to vanilla Chatterbox recovers commercial-class Hindi intelligibility -giving us a two-branch deployment (LoRA branch for Te/Ta, vanilla branch for Hi) with a shared inference recipe. Evaluated on 10-utterance pilot sets with the companion PSP [1] benchmark, our Praxy Voice system is on par with or slightly ahead of the best commercial baselines on three per-language headline metrics: 26.7% retroflex collapse on Telugu (vs Sarvam Bulbul's 33.3%; within the n = 15-token noise band but consistently ranked first), 71% Tamil-zha collapse (vs the commercial trio's 86%; the single clearest perdimension gain we observe), and 0.025 LLM-WER on Hindi (tied with Cartesia Sonic-3). All of this from a Chatterbox base that natively did not cover Telugu or Tamil at all, using zero commercial TTS training data. Finally, for intra-sentential code-mixed input we add a third branch (IndicF5 + native-script transliteration preprocessor) that drops LLM-WER on code-mix from 0.80-0.85 (raw IndicF5) to 0.14-0.27 across Hi/Te/Ta -closing most of the gap on Te while remaining clearly behind commercial systems on Hi. We release the R6 LoRA weights under Apache-2.0, the inference code (including the code-mix preprocessor and a unified production router) under MIT, and a Gradio demo.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2604.25441")
get_code_for_paper("2604.25441")
have("2604.25441")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2604.25441.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2604.25441.svg)](https://syntology.ai/paper/2604.25441)

Connect an agent — have() is free.