xby-asr-f

Most speech-to-text engines are tuned for standard Mandarin and degrade sharply—or fail entirely—when fed Cantonese, Sichuan-Chongqing, Northeastern, Minnan, or other regional dialects. xby-asr-f closes that gap by wiring the Fire-Red ASR CTC model (provided by Xiaoben Yang) directly into the DeepSeek Harness pipeline, so dialect-aware transcription happens inside your existing conversation loop without juggling external APIs.

The plugin offers three flexible input paths: an audio URL, a Base64-encoded payload, or a local file path, covering everything from streamed media to recorded files on disk. Your API key is set once in chat and persisted automatically, surviving session restarts with no re-entry. Once transcription completes, the text result can be chained straight into follow-up prompts for translation, summarization, or downstream NLP tasks—all within the same DSH thread.

It is a good fit for NLP developers working with multi-accent voice data, teams building dialect-aware chatbots or voice assistants, and researchers whose work involves regional language varieties. Shipped under the MIT license, it can be freely integrated into both commercial and open-source projects.