Open Issues Need Help
View All on GitHubAI Summary: The FluidAudio project fails to build with Swift 6 strict concurrency enabled due to data race errors in `StreamingAsrManager.swift`. The `StreamingAsrManager` actor stores a `nonisolated(unsafe)` reference to `AsrManager`, and when a local binding of this reference is used across an `await` boundary, Swift 6 flags it as a potential data race because `AsrManager` is likely not `Sendable`.
Frontier CoreML audio models in your apps — text-to-speech, speech-to-text, voice activity detection, and speaker diarization. In Swift, powered by SOTA open source.
Frontier CoreML audio models in your apps — text-to-speech, speech-to-text, voice activity detection, and speaker diarization. In Swift, powered by SOTA open source.
Frontier CoreML audio models in your apps — text-to-speech, speech-to-text, voice activity detection, and speaker diarization. In Swift, powered by SOTA open source.
Frontier CoreML audio models in your apps — text-to-speech, speech-to-text, voice activity detection, and speaker diarization. In Swift, powered by SOTA open source.
AI Summary: The task is to debug and fix a bug in the `AsrModels.load` function within the FluidAudio Swift framework. The function currently ignores user-specified compute units (`configuration.computeUnits`) and instead always uses optimized compute units (`optimizedConfig.computeUnits`). The fix requires modifying the `load` function to correctly utilize the user-provided compute unit configuration.
Frontier CoreML audio models in your apps — text-to-speech, speech-to-text, voice activity detection, and speaker diarization. In Swift, powered by SOTA open source.
AI Summary: The task is to provide a Swift code snippet demonstrating live Automatic Speech Recognition (ASR) from a microphone on iOS using the FluidAudio library. The snippet should handle audio input from the microphone, process it using the library's (currently missing) `transcribeChunk` function (or an equivalent), and output the transcription to the console. The solution needs to account for the user's inexperience with iOS development.
Frontier CoreML audio models in your apps — text-to-speech, speech-to-text, voice activity detection, and speaker diarization. In Swift, powered by SOTA open source.
AI Summary: The task is to modify the FluidAudio Swift framework to prevent the repeated download of VAD models from HuggingFace upon each app launch. This involves integrating the pre-downloaded CoreML models directly into the Xcode project, eliminating the need for online retrieval.
Frontier CoreML audio models in your apps — text-to-speech, speech-to-text, voice activity detection, and speaker diarization. In Swift, powered by SOTA open source.