零成本让非印地语基模型实现商用级印地语语音合成
Praxy Voice: Voice-Prompt Recovery + BUPS for Commercial-Class Indic TTS from a Frozen Non-Indic Base at Zero Commercial-Training-Data Cost
- 用统一音素空间将7种印地文字转为拉丁字符，让现有模型处理印地语
- 仅训练文本预测器的LoRA适配器，1220小时授权数据即达商用效果
- 语音提示恢复技术无需训练声学解码器，8-11秒参考音频即可出效果
商用语音合成系统可生成接近母语者的印地语语音，但最佳开源基模型（Chatterbox、Indic Parler-TTS、IndicF5）在发音维度上仍有差距，且广泛使用的多语言基模型（Chatterbox，支持23种语言）不支持泰卢固语和泰米尔语。本文提出：如何在不训练新声学解码器、不使用任何商用语音训练数据的前提下，使非印地语基模型达到商用级印地语语音合成效果？解决方案包括：（1）BUPS——一种确定性将7种印地文字转换为ISO-15919拉丁转写的婆罗米统一音素空间，使Chatterbox的拉丁分词器可处理；（2）仅对文本分词器（Chatterbox的t3）训练一个LoRA适配器，使用约1220小时授权印地语音频及印地语代理语言标识符；（3）语音提示恢复方案——使用8-11秒同语言参考片段，配合三重采样参数（夸张系数0.7，温度0.6，最小概率0.1；“Config B”），无需声学解码器训练即可恢复商用级输出。在10句样本集上的评测显示，Praxy Voice在泰卢固语上后置音塌陷率26.7%（优于Sarvam Bulbul的33.3%），泰米尔语-zha塌陷率71%（优于商用三强的86%），印地语LLM-WER为0.025（与Cartesia Sonic-3并列）。针对句内代码混杂场景，新增第三分支（IndicF5 + 原文转写），使代码混杂的LLM-WER从0.80–0.85降至0.14–0.27。项目已发布R6 LoRA权重（Apache-2.0）、推理代码与路由器（MIT），以及Gradio演示。
原文摘要 · Abstract (English)
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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。