voice-isolator

Removes background noise and isolates vocals or speech from audio using the ElevenLabs Voice Isolator (audio isolation) API. It suits cleaning up noisy recordings, stripping music or ambient sound from dialogue, isolating speech from field recordings, preparing audio for transcription, or extracting vocals from a mixed track. The skill requires internet access and an ElevenLabs API key supplied via the ELEVENLABS_API_KEY environment variable.

It provides quick-start examples in Python, JavaScript, and cURL. The Python and JavaScript SDKs call audio_isolation.convert (audioIsolation.convert in JS) with an audio file and stream the cleaned result to disk; the cURL example posts to the v1/audio-isolation endpoint with the xi-api-key header. There are only two documented parameters: the required audio file, and an optional file_format that defaults to other for any encoded audio, or pcm_s16le_16 for 16-bit PCM mono at 16kHz little-endian, which skips decoding to reduce latency. Input accepts any common encoded audio or video container (MP3, WAV, M4A, FLAC, OGG, WebM, MP4, and others), and the response is a streamed MP3 by default. It also shows how to isolate audio fetched from a URL by downloading it into a BytesIO buffer first.

Use it whenever a task involves denoising or voice isolation as a preprocessing step or a standalone cleanup. Documented workflows include cleaning up interview and podcast recordings, isolating voice before speech-to-text for better transcription accuracy, extracting dialogue from tracks mixed with music or SFX, and pre-processing a source voice before applying a voice changer. The skill includes basic error handling and lists common API errors: 401 for an invalid API key, 422 for invalid parameters such as a wrong file_format for the supplied audio, and 429 for rate limiting. A separate installation guide is referenced for setup.