Common Issues

Nothing is transcribed

• Open Settings → Permissions. Microphone Access is marked Required — if it is not granted, click Grant Access.
• Check the input device in Settings → General → Microphone and click Change to pick another one.
• Check the engine in Settings → Transcription. With Local Model (On-Device) the selected model must be labelled Downloaded; with a BYOK provider its API key must be saved.
• Raise Microphone sensitivity in Settings → System and press its Test button — it records three seconds and plays them back at your current gain.

The shortcut does nothing

• Confirm Speakly is running: its icon is in the menu bar / tray and its menu opens.
• On macOS, grant Accessibility Access in Settings → Permissions. The shortcut dialog says so itself: "Make sure to grant Accessibility permissions if prompted."
• Change the combination if another app already owns it: Settings → General → Keyboard shortcut → View, then Change Shortcut in the Keyboard Shortcut dialog. Press the keys you want — the hint reads "Press keys to add them, press again to remove" — then Done.
• If Speakly answers Could not set that shortcut. Try a different combination., the combination is unavailable; pick another.

Poor transcription quality

• Speak at a steady volume and cut background noise.
• Try a larger model in Settings → Transcription — for example large-v3 instead of a quantized turbo model.
• Add names and technical terms in the sidebar's Dictionary page. See Custom Dictionary.
• If capitalisation and punctuation are the problem rather than the words, that is Style, not accuracy — see Writing Style.

FAQ or flashcards fail on a long video

This used to end in one sentence whatever the cause: "Failed to generate content. Please try again." The message now carries the provider's own error and the size that produced it, in the form <provider>: <the provider's message> (sent <n> characters) — a rate limit and a bad key no longer look the same.

• Long transcripts are split at 16,000 characters on sentence boundaries, processed one part after another, and the parts are joined. Sequential on purpose: parts sent at once would compete for the same per-minute budget.
• A rate-limited request is no longer treated as a failure. With Groq, Speakly waits the delay Groq names in the response and retries, at most twice, never waiting longer than 65 seconds at a time.
• A request refused as too large is not retried — waiting cannot make one request fit inside a per-minute budget it already exceeds. Splitting is what fixes that, and it is now automatic.
• If it still fails, the text you see is the provider's, not Speakly's. Read it before retrying: it names the limit and the size.