Show HN: SpeechAnalyzerDylib – C-FFI for macOS 26.0 SpeechAnalyzer API

4 hours ago 2

Starting with macOS Tahoe, the SpeechAnalyzer API will be introduced. To make it easier to use, I’ve set up an environment to wrap it into a dylib.

File Transcription Call sw_transcribeFile(const char *filePath, const char *locale, TranscriptionCallback callback, void *userData) to transcribe an audio file asynchronously.
- Inputs:
  - filePath: C-string path to the audio file
  - locale: C-string locale identifier (defaults to current locale if NULL)
  - callback: C function of type void (*)(const char *textOrError, void *userData)
  - userData: opaque pointer passed back in each callback
- Behavior:
  - Launches a background task to read and process the file
  - On success, sends the full transcript to your callback
  - On failure, sends an "Error: …" string instead
- Return: 0 if the task was started, -1 if no file path was provided
Data Transcription Call sw_transcribeData(const uint8_t *bytes, size_t size, const char *locale, TranscriptionCallback callback, void *userData) to transcribe in‐memory audio data.
- Inputs:
  - bytes: pointer to the raw audio data buffer
  - size: number of bytes in the audio buffer
  - locale: C‐string locale identifier (defaults to current locale if NULL)
  - callback: C function of type void (*)(const char *textOrError, void *userData)
  - userData: opaque pointer passed back in each callback
- Behavior:
  - Copies the buffer into a temporary file on disk
  - Reuses the existing file‐based transcription path (transcribeFile)
  - On success, sends the full transcript to your callback
  - On failure, sends an "Error: …" string instead
  - Cleans up the temporary file when done
- Return: 0 if the task was started, -1 if the buffer pointer is NULL or size is zero
Live Microphone Transcription Call sw_startMicrophoneTranscription(const char *locale, TranscriptionCallback callback, void *userData) to begin streaming live speech from the default mic.
- Inputs: same as above, except no file path
- Behavior:
  - Verifies or installs the offline model for the given locale
  - Captures, converts, and analyzes mic audio in real time
  - Streams each partial transcript back via your callback as it arrives
  - Stops and finalizes automatically when the audio engine is stopped
- Return: always 0 (task is fire-and-forget)