Abstract

Some real-time speech translation systems are architected for monolingual input and can encounter challenges, such as transcription errors or increased latency, when processing code-switched speech. The disclosed method describes a system for real-time interpretation that can ingest a raw audio stream and utilize a dual-axis model to concurrently identify both a spoken word and its language of origin. This process can produce a token stream with associated language identifiers. A subsequent syntactic bridge module can analyze this mixed-language stream to generate a language-independent grammatical representation of the utterance's semantic content. This abstract structure may then be used to synthesize a sentence in a desired target language, facilitating the translation of code-switched conversation.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Share

COinS