Set up SayMyWork
1. Choose transcription and connect OpenAI
Custom transcription service
Choose Custom compatible service in Setup. Enter the full transcription URL, model name and optional service key. The service must accept an OpenAI-compatible file transcription request and return JSON with a text field. HTTPS is required for remote services; HTTP is allowed on localhost. Audio goes directly to your chosen service, whose fees and privacy policy apply. Use Test custom transcription to check the selected microphone and meeting output. Other API formats need a dedicated adapter.
Local Whisper
Use Windows 11, a compatible 64-bit processor and the Microsoft Visual C++ 2022 x64 runtime. A compatible GPU can speed up recognition; performance varies by PC.
Download the English Whisper model (approximately 148 MB). In Setup, choose Local Whisper, click Choose model and select the downloaded file. Audio is transcribed on your PC without Azure credits. Other languages need a compatible multilingual model. See the current update and limits.
OpenAI answers
Create an OpenAI API project, fund API usage and add its key in the app. Start with the default small model. OpenAI API keys · OpenAI pricing.
If you prefer Azure transcription, create an Azure Speech S0 resource and enter its key and matching region. The app uses two concurrent transcription streams. F0’s documented one-request concurrency is unsuitable for this architecture against one resource. Azure portal · Speech quotas.
2. Prepare truthful notes
Add your résumé, role description and project notes under the correct source types. Include your contribution, decisions, failure recovery and actual results. Add a speaking sample for style. Check extraction before generating.
3. Check audio locally
Use headphones. Match Teams’ speaker and microphone to the app’s selected devices. Play a Teams speaker test and speak during Check audio. Both meters should respond separately. The level check does not call Azure. In Local Whisper mode, also choose Test local transcription, start its 20-second check, wait for the beep, then speak and play meeting speech. Review the recognized wording as well as the source counts. The recognition check does not call Azure or OpenAI.
4. Start your one session deliberately
Verify your email in Trial / license. You can use typed rehearsal without speech capture. For a live meeting, obtain permission and disclose capture first. Warnings appear at ten and two minutes remaining.
5. Review and retain
Check generated claims against the sources, edit wording and mark prepared entries reviewed. Changing source material invalidates prior review. Save or export your work. Stop & review ends capture.
Beta limits
Teams must pass the live launch gate. Zoom, Google Meet, Webex and Other remain beta until tested separately. Source citations do not guarantee a factual sentence is correct. The app cannot guarantee interview results.