🎛️ We Customized Translation, Why Not Interpreting?
CCMI is a desktop PyQt5 app that turns your microphone into a customizable consecutive interpreter: mic → Whisper ASR → GPT (brief + termlist + context) → TTS. It adapts to the session (one‑way, two‑party, or two‑party + audience), keeps terminology consistent, and speaks back in the voice you choose.
- Windows (ready-to-use):Download CCMI_Windows_x64.zip
- Or visit the latest release for all assets.
After download: unzip and run
ccmi.exe. Click 🔑 Set API Key and paste an OpenAI key (sk-…).
- 🧵 Customization doesn’t cross the mic. Traditional tools treat interpreting as one-size-fits-all: no pre‑briefs, no termlists, no tone or audience intent.
- ⏱️ Too many steps, too much lag. Speech → text → translation → voice. Each hop adds delay and loses detail.
- 🧩 Sessions are not identical. A sales call, a lecture, and a panel get treated the same. No session modes or memory per party.
- 🧭 Fits your setup. Solo talk? Two-person call? Conversation with listeners? Pick the right mode so roles and direction are clear.
- 🎙️ Tell it once, CCMI prepares. Describe your session and CCMI fills a minimal brief: purpose, roles, tone, and rules.
- 📚 Your terms, locked in. Import a list or add your own. Names and phrases stay consistent across the whole session.
- 🎧 Voices for every tone. Choose among built-in voice styles and test them anytime.
- 🧠 Context that grows. Each new segment adapts to your brief, prior translations, and termlist.
Session selection
Two-party customization
Translation & review
- Session modes: One‑Way, Two‑Party, Two‑Party + Audience
- Briefs for each direction, plus Audience fields when relevant
- Termlist: CSV/XLSX import, in‑app edit, clear/add/delete
- Review table: “Copy Last” + XLSX export
- Device picker & meters, language Swap (Ctrl/⌘+Enter), Shift+Space to record
- Voice tester and multiple TTS styles
- Privacy: API key lives in memory only; temp audio files are removed
- Online: uses OpenAI APIs for ASR, translation, and TTS
CCMI uses OpenAI for the full pipeline:
- ASR (speech → text):
whisper-1 - Translation / brief filling: default
gpt-4.1-2025-04-14(editable in the UI) - TTS (text → speech):
gpt-4o-mini-tts
Your API key is stored in memory only for the running session. Temporary audio files are cleaned up after use.
| Feel | Voice ID |
|---|---|
| neutral, balanced | alloy |
| warm, natural | ash |
| deep, rich | ballad |
| bright, expressive | coral |
| clear, energetic | echo |
| calm, steady | sage |
| light, youthful | shimmer |
| bold, dramatic | verse |
In two-party modes, CCMI automatically picks the section’s voice based on direction. Use 🧪 Hear to audition.
Requirements: Python 3.9+, microphone & speakers.
python -m venv .venv
# Windows
.venv\Scripts\activate
# macOS / Linuxsource .venv/bin/activate
pip install -r requirements.txt
python ccmi.py| Action | Keys |
|---|---|
| Start/Stop recording | Shift + Space |
| Swap Source ↔ Target | Ctrl + Enter (Win/Linux) / ⌘ + Enter (macOS) |
| Test section voice | 🧪 Hear button |
- 🎯 One‑Way (A → B) — announcements/lectures; includes Audience fields.
- 🤝 Two‑Party (A ↔ B) — conversations/calls; separate briefs for each direction.
- 🎤 Two‑Party + Audience (A ↔ B + 👥) — conversations with listeners; adds Audience fields.
- Two columns: Source Term, Target Term (headers optional).
- Import via Termlist → “Import Termlist (XLSX/CSV)”.
- During translation, exact pairs are enforced (
source = target). (XLSX import/export requiresopenpyxl— optional dependency.)
- Review → Export XLSX saves the segment table with wrapped cells.
- Copy Last puts the newest translation on your clipboard.
This repo ships with ccmi.spec for PyInstaller.
pip install -r requirements.txt
pip install pyinstaller
pyinstaller ccmi.spec- Your app will be in
dist/ccmi/. Zip that folder and attach it as a Release asset (don’t commitdist/orbuild/). - To shrink size, install UPX and pass
--upx-dirto PyInstaller, or remove optional deps you don’t need (e.g.,openpyxl).
- GUI: PyQt5
- Audio I/O: sounddevice, soundfile (16 kHz mono capture)
- ASR: OpenAI
whisper-1 - Translation: Chat Completions (model configurable; default
gpt-4.1-2025-04-14) - TTS:
gpt-4o-mini-ttswith selectable voices (alloy,ash,ballad,coral,echo,sage,shimmer,verse)
- API key is requested at runtime and not written to disk.
- Temporary audio files are deleted after use.
- No telemetry, analytics, or background upload.
- You are responsible for complying with OpenAI API Terms in your region.
- “No OpenAI API key set.” → Click Set API Key.
- “No usable audio devices were found.” → Plug in a mic/speaker and reopen CCMI.
- Import XLSX fails → Install
openpyxlor use CSV instead. - No sound on playback → Pick the correct 🔊 output device and try 🧪 Hear.
- Large prompts → CCMI caps rolling context to ~5,000 chars to keep latency reasonable.
PRs and issues are welcome! If you’re proposing larger changes, please open an issue first to discuss scope/fit.
MIT — see LICENSE.