Add inputDeviceID passthrough to AudioStreamTranscriber

[DTM-7-RX]

Summary

Adds an optional inputDeviceID parameter to AudioStreamTranscriber and forwards it to the existing AudioProcessor.startRecordingLive(inputDeviceID:) API.

Today, streaming transcription always records from the system default input because AudioStreamTranscriber calls startRecordingLive() without passing a device. Lower-level recording already supports a specific input device, but the streaming wrapper does not expose it.

This is additive and keeps existing behavior unchanged when inputDeviceID is nil.

Why

This unblocks macOS apps that let users select an input device for streaming transcription, such as virtual or aggregate devices. For example, routing meeting audio into BlackHole works for non-streaming recording, but streaming transcription currently ignores the selected BlackHole device and captures the system default input instead.

I verified this fix in a local downstream app by routing audio into BlackHole and passing its device ID through AudioStreamTranscriber; streaming transcription then captured the selected device as expected.

Fixes #493.

Testing

• git diff --check HEAD~1..HEAD
• make build

Note: make build required running outside the Codex sandbox because SwiftPM's own sandbox-exec fails with sandbox_apply: Operation not permitted inside the Codex sandbox. The build completed successfully outside that sandbox.

[DTM-7-RX]

Hi @ZachNagengast, just a gentle follow-up on this PR.

I confirmed that the issue is still present in v1.0.0 and the current main branch. We have also validated this patch in a downstream macOS app: streaming transcription correctly captures from BlackHole while the system default input remains the MacBook microphone.

Would you be open to reviewing this? I’m happy to add tests or adjust the API if needed. I also have a separate mitigation for the input-format/installTap crash, which I intentionally kept out of this focused PR.

[ZachNagengast]

Hi @DTM-7-RX Could you merge main in here and validate against that with latest changes? Looks like a small contained change, if tests pass will be ok to merge as is from what I can tell.

[DTM-7-RX]

Thanks @ZachNagengast! Merged main in — no conflicts.

Validated on the merged branch (Swift 6.3.2, arm64 macOS):

• swift build — clean
• AudioProcessorTests — 12 tests, 0 failures
• WhisperKitTests.UnitTests — 131 tests, 1 skipped, 0 failures

No CI checks have run on the new commit yet — could you kick those off?

[Copilot]

Pull request overview

Adds an optional inputDeviceID parameter to AudioStreamTranscriber so streaming transcription can record from a caller-selected input device (instead of always using the system default), by forwarding through to the existing AudioProcessor.startRecordingLive(inputDeviceID:) API.

Changes:

• Added inputDeviceID: DeviceID? = nil to AudioStreamTranscriber initializer and stored it as a private property.
• Passed inputDeviceID through when starting live recording in startStreamTranscription().

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.

Suppressed comments (1)

Sources/WhisperKit/Core/Audio/AudioStreamTranscriber.swift:45

• The new inputDeviceID comment reads as if device selection is universally supported. In this codebase DeviceID is only meaningful on macOS (elsewhere it’s a String and recording always uses the default input), so this comment may mislead future maintainers/callers reading the source.

    /// Optional input device to capture from. nil = system default input.
    /// Lets callers stream from a specific device (e.g. a virtual device) instead
    /// of always using the system default. Passed through to startRecordingLive.
    private let inputDeviceID: DeviceID?

[ZachNagengast]

Nice improvement, thanks for the contribution @DTM-7-RX!