Transcription providers
Choose a local engine, a configured transcription server, or a cloud provider. Each option sends audio to its selected destination.
At a glance
Choose and configure a transcription provider in Settings > Transcription. It processes audio for new recordings. To choose the model that turns the finished transcript into a meeting summary, see Summary providers.
Transcription providers
Meetily uses one selected transcription provider for new recordings. Choose an option that matches where you want audio processed and how you want to manage its model or credentials.
Choose a transcription provider
Where audio goes
Meetily sends audio to the transcription engine you select. Whisper and Parakeet process it on this device. OpenAI-Compatible sends it to the URL you configure, while Groq and OpenAI send it to that provider's service.
On this device
| Provider | Use it when | Configuration |
|---|---|---|
| Whisper | You use macOS or Windows with a supported GPU and want the quality-focused local option, broader language coverage, automatic detection, and a manual language choice. | Download and select a model. |
| Parakeet | You use Windows without a supported GPU and want local CPU transcription. It supports English plus 24 European languages, but has a narrower language range and no manual language choice. | Download the model when prompted. |
Configured server
| Provider | Use it when | Configuration |
|---|---|---|
| OpenAI-Compatible | You run or have access to a server that supports the OpenAI transcription API. | Provide its server URL, and an API key if that server requires one. |
Provider cloud
| Provider | Use it when | Configuration |
|---|---|---|
| Groq | You want to use Groq's transcription service. | Provide a current Groq API key and select an available model. |
| OpenAI | You want to use OpenAI's transcription service. | Provide a current OpenAI API key and select an available model. |
Choose based on your device and language needs
Use Whisper on macOS or Windows with a supported GPU when quality, language coverage, automatic detection, or language selection matter most. Use Parakeet on Windows without a supported GPU when its supported language set meets your needs.
How to switch providers
- Open Settings > Transcription.
- Select a provider in Transcription Provider.
- Complete the configuration that appears for that provider.
- Select a model where the provider offers a model list.
- For a server or cloud provider, use Test Connection before recording.
The provider and model controls are unavailable while a recording is in progress. Finish the recording before changing them.
Provider settings are stored separately, so returning to a provider restores its saved configuration.
Set up each provider
Whisper (local)
Whisper is the quality-focused local option for macOS and Windows devices with a supported GPU. It supports a broader range of languages than Parakeet, detects the spoken language automatically, and lets you choose a language when needed. The basic model list includes small, medium-q5_0, large-v3-q5_0, large-v3-turbo, and large-v3; more models are available under Advanced.
- Select Whisper in Settings > Transcription.
- In the model manager, download a model from the basic list or Advanced.
- Select a downloaded model.
- Choose a language when needed, then start recording.
Add a custom Whisper model
Use this only when a compatible model repository provides a model you need.
- Open the Whisper model manager and select Add from HuggingFace.
- Enter the repository URL or repository path.
- Choose a compatible model file from the available files.
- Download it, then select it in the model manager when the download completes.
If a repository requires authentication, use a Hugging Face token with access to that repository. Meetily validates selected model files before using them.
Parakeet (local)
Parakeet is the practical local option for Windows devices without a supported GPU and runs on CPU. Its current v3 model automatically selects the spoken language from these supported languages: Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Greek, Hungarian, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Slovak, Slovenian, Spanish, Swedish, Russian, and Ukrainian.
Meetily does not show a detected-language result for Parakeet, and Parakeet does not provide a manual language selector. The optional v2 model is English-only.
- Select Parakeet in Settings > Transcription.
- Download the model when Meetily prompts you.
- Start recording after the model is ready.
OpenAI-Compatible
Choose this option for a server that accepts the OpenAI transcription API. The destination can be a server on this device, a server on your network, or an internet-hosted service.
- Select OpenAI-Compatible in Settings > Transcription.
- Enter the server URL.
- Add an API key if that server requires one.
- Wait for its models to load, then select a model.
- Select Test Connection before recording.
Groq and OpenAI
Choose one of these providers when you want to send audio to its transcription service.
API keys are saved locally and masked in the app. When used, a key is sent only to the provider you configured.
- Select Groq or OpenAI in Settings > Transcription.
- Enter the API key for that provider.
- Choose an available model after the model list loads.
- Select Test Connection before recording.
Connection testing
For OpenAI-Compatible, Groq, and OpenAI, test the connection before the first recording. A successful test confirms that Meetily can reach the selected service and use the selected model. If it fails, check the server URL, credentials, service availability, and selected model.
During recording
Meetily captures audio and sends it to the selected transcription engine. For Whisper and Parakeet, processing stays on this device. For OpenAI-Compatible, Groq, and OpenAI, audio is sent to the configured server or provider service.
If a request to an external provider fails, the audio recording remains saved locally. Address the provider configuration, then retranscribe the saved audio.
Frequently asked questions
No. One provider is active at a time for new recordings. Change it before starting a recording.
OpenAI is OpenAI's provider service. OpenAI-Compatible is a server that accepts the same transcription API and can run on this device, on your network, or on the internet.
Changing transcription engines during a recording could create an inconsistent transcript. Finish the recording, then change the provider or model.
Last updated on