v0.1.0
CrossMeet v0.1.0 · Internal dev milestone
The first internal dev build — bidirectional simul-interpret, local + cloud ASR, virtual mic, glossary, knowledge base RAG.
NEW
Architecture baseline
After 6 months of development, CrossMeet shipped its first internal dev build (Windows 10 / 11) — the architecture baseline that became v1.0. Highlights:
Real-time translation
- Bidirectional simul-interpret pipeline: microphone + system audio run independent ASR, left/right panels render separately
- 30+ languages, any pair, auto-detected
- Iron rule #1 enforced: source text arrives in ~100 ms, translation streams in asynchronously and back-fills — never blocks
ASR engines
- Local Whisper-faster (GPU/CPU)
- Whisper API (OpenAI-compatible)
- Aliyun DashScope ASR
- Qwen3-ASR (low-latency Chinese)
Translation / TTS
- LLM: OpenAI / Anthropic / Ollama / local LLM
- TTS: Edge / OpenAI / CosyVoice / VoxCPM / IndexTTS2 / Aliyun
Pro features
- Glossary: 3 types × group activation
- Knowledge base (RAG): PDF/DOCX vector retrieval injected into prompt
- Compose bar (interpret mode): type → preview → correct → send 100% accurate
- Virtual mic: Zoom / Teams / Discord hear native-quality speech directly
- History + export: session recording + Markdown / PDF / DOCX