henrisel/DeltaVR3DModelGeneration
1
0
Fork 0
forked from cgvr/DeltaVR
Code Pull Requests Activity

improve fmod whisper bridge and radio button using it

Browse Source
This commit is contained in:
henrisel
2026-02-21 20:25:12 +02:00
parent c968e6bed4
commit 1a29f785b8
4 changed files with 204 additions and 86 deletions
Download Patch File Download Diff File Expand all files Collapse all files

270
Assets/_PROJECT/Scripts/ModeGeneration/FMODWhisperBridge.cs
View File

@@ -5,18 +5,19 @@ using UnityEngine;
using FMOD;
using FMODUnity;
using Whisper; // WhisperManager, WhisperStream, WhisperResult
using Whisper.Utils; // AudioChunk
using Whisper.Utils;
using Debug = UnityEngine.Debug; // AudioChunk
/// <summary>
/// Capture microphone with FMOD and feed chunks to WhisperStream (no Unity Microphone).
/// Also (optionally) plays the recorded sound back via FMOD loopback.
/// FMOD mic is initialized once (Start) and runs continuously in a ring buffer.
/// Whisper stream is created when ActivateRecording() is called and disposed on DeactivateRecording().
/// Optional loopback can be paused/resumed instead of starting/stopping to avoid stalls.
/// </summary>
public class FMODWhisperBridge : MonoBehaviour
{
[Header("Whisper")]
[SerializeField] private WhisperManager whisper; // assign in Inspector
[SerializeField] private bool useVadInStream = false; // let WhisperStream do VAD or not
[SerializeField] private bool useVadInStream = true; // let WhisperStream do VAD or not
[Header("FMOD capture")]
[Tooltip("Recording device index (0 = default)")]
@@ -29,6 +30,8 @@ public class FMODWhisperBridge : MonoBehaviour
[Header("Loopback (monitor your voice)")]
public bool playLoopback = true;
[Tooltip("If true, loopback plays only while active; otherwise it’s always on.")]
public bool loopbackOnlyWhenActive = true;
[Range(0f, 2f)] public float loopbackVolume = 1.0f;
public delegate void OnWhisperSegmentUpdatedDelegate(string result);
@@ -55,7 +58,9 @@ public class FMODWhisperBridge : MonoBehaviour
// temp conversion buffer
private float[] _floatTmp = new float[0];
private short[] _shortOverlay;
// activation flag
private bool isRecordingActivated = false;
private void Awake()
@@ -66,85 +71,180 @@ public class FMODWhisperBridge : MonoBehaviour
private async void Start()
{
// Query device info to get native rate/channels.
// (FMOD: getRecordDriverInfo gives you system rate & speaker mode)
// -------------- FMOD initialize ONCE --------------
// Query device info
string name;
Guid guid;
SPEAKERMODE sm;
int smChannels;
DRIVER_STATE driverState;
// signature: getRecordDriverInfo(id, out name, nameLen, out guid, out systemrate, out speakermode, out speakermodechannels, out driverState)
_core.getRecordDriverInfo(recordDriverId, out name, 256, out guid, out _nativeRate, out sm, out smChannels, out driverState);
_nativeChannels = channels > 0 ? channels : smChannels;
UnityEngine.Debug.Log($"[FMOD→Whisper] Using input device #{recordDriverId}: \"{name}\" rate={_nativeRate} ch={_nativeChannels}");
// Build a user sound buffer that FMOD will fill (OPENUSER | LOOP_NORMAL).
var res = _core.getRecordDriverInfo(
recordDriverId,
out name, 256,
out guid,
out _nativeRate,
out sm,
out smChannels,
out driverState
);
if (res != RESULT.OK)
{
Debug.LogError($"[FMOD→Whisper] getRecordDriverInfo failed: {res}");
return;
}
_nativeChannels = channels > 0 ? channels : smChannels;
int rate = (_nativeRate > 0) ? _nativeRate : desiredSampleRate;
Debug.Log($"[FMOD→Whisper] Using input device #{recordDriverId}: \"{name}\" rate={rate} ch={_nativeChannels}");
// Build user sound (ring buffer) — multiple seconds
CREATESOUNDEXINFO ex = new CREATESOUNDEXINFO
{
cbsize = Marshal.SizeOf(typeof(CREATESOUNDEXINFO)),
numchannels = _nativeChannels,
defaultfrequency = (_nativeRate > 0) ? _nativeRate : desiredSampleRate,
defaultfrequency = rate,
format = SOUND_FORMAT.PCM16,
length = (uint)(((_nativeRate > 0 ? _nativeRate : desiredSampleRate) * _nativeChannels) * sizeof(short)) // seconds=1 (we loop)
length = (uint)(rate * _nativeChannels * sizeof(short) * bufferLengthSec)
};
_core.createSound("", MODE.OPENUSER | MODE.LOOP_NORMAL | MODE.CREATESAMPLE, ref ex, out _recSound);
res = _core.createSound("", MODE.OPENUSER | MODE.LOOP_NORMAL | MODE.CREATESAMPLE, ref ex, out _recSound);
if (res != RESULT.OK)
{
Debug.LogError($"[FMOD→Whisper] createSound failed: {res}");
return;
}
_recSound.getLength(out _soundPcmLength, TIMEUNIT.PCM);
// Start FMOD recording into that sound (looping ring buffer).
_core.recordStart(recordDriverId, _recSound, true);
UnityEngine.Debug.Log("[FMOD→Whisper] Recording started.");
// Start recording (looping)
res = _core.recordStart(recordDriverId, _recSound, true);
if (res != RESULT.OK)
{
Debug.LogError($"[FMOD→Whisper] recordStart failed: {res}");
_recSound.release();
_recSound.clearHandle();
return;
}
// Optional loopback playback using FMOD (plays same sound ring buffer).
// Initialize record position to avoid a huge first delta
_core.getRecordPosition(recordDriverId, out _lastRecordPos);
Debug.Log("[FMOD→Whisper] Recording started.");
// Loopback channel (optional). Start once; pause when inactive if desired.
_core.getMasterChannelGroup(out _masterGroup);
if (playLoopback)
{
_core.playSound(_recSound, _masterGroup, false, out _playChannel);
_playChannel.setMode(MODE._2D);
_playChannel.setVolume(loopbackVolume);
UnityEngine.Debug.Log("[FMOD→Whisper] Loopback playback started.");
res = _core.playSound(_recSound, _masterGroup, false, out _playChannel);
if (res == RESULT.OK && _playChannel.hasHandle())
{
_playChannel.setMode(MODE._2D);
_playChannel.setVolume(loopbackVolume);
if (loopbackOnlyWhenActive) _playChannel.setPaused(true); // keep muted until Activate
Debug.Log("[FMOD→Whisper] Loopback playback ready.");
}
else
{
Debug.LogWarning($"[FMOD→Whisper] playSound failed or channel invalid: {res}");
}
}
// Create Whisper stream WITHOUT MicrophoneRecord, just from (freq, channels).
// We'll push AudioChunk manually.
// NOTE: WhisperStream’s sliding window is governed by manager’s stepSec/keepSec/lengthSec.
_stream = await whisper.CreateStream(ex.defaultfrequency, _nativeChannels);
_stream.OnResultUpdated += (txt) =>
// No Whisper stream here. It will be created on ActivateRecording().
await System.Threading.Tasks.Task.Yield();
}
/// <summary>
/// Creates a fresh Whisper stream and starts feeding audio (FMOD already recording).
/// </summary>
public async void ActivateRecording()
{
if (isRecordingActivated)
{
//OnWhisperResultProcessed?.Invoke(txt);
};
Debug.Log("[FMOD→Whisper] ActivateRecording ignored (already active).");
return;
}
if (!_recSound.hasHandle())
{
Debug.LogError("[FMOD→Whisper] FMOD not initialized or recording not running.");
return;
}
int rate = (_nativeRate > 0) ? _nativeRate : desiredSampleRate;
try
{
_stream = await whisper.CreateStream(rate, _nativeChannels);
}
catch (Exception e)
{
Debug.LogError($"[FMOD→Whisper] CreateStream exception: {e}");
_stream = null;
_streamStarted = false;
return;
}
// Wire events
_stream.OnSegmentUpdated += (seg) =>
{
if (IsSpeechMeaningful(seg.Result))
{
OnWhisperSegmentUpdated?.Invoke(seg.Result);
}
};
_stream.OnSegmentFinished += (seg) =>
{
if (IsSpeechMeaningful(seg.Result))
{
OnWhisperSegmentFinished?.Invoke(seg.Result);
}
};
// If you want Whisper to respect VAD, enable in manager or set useVad (manager controls stream params).
whisper.useVad = useVadInStream;
_stream.StartStream();
_streamStarted = true;
// prepare temp arrays roughly 100ms of audio
EnsureTmpCapacity((ex.defaultfrequency / 10) * _nativeChannels);
// Unpause loopback if it's meant to play only while active
if (playLoopback && loopbackOnlyWhenActive && _playChannel.hasHandle())
_playChannel.setPaused(false);
// Prepare temp arrays roughly 100ms of audio
EnsureTmpCapacity((rate / 10) * _nativeChannels);
isRecordingActivated = true;
Debug.Log("[FMOD→Whisper] Stream activated (Whisper started; FMOD was already recording).");
}
/// <summary>
/// Stops and disposes the Whisper stream only. FMOD keeps recording.
/// </summary>
public void DeactivateRecording()
{
if (!isRecordingActivated && !_streamStarted)
return;
isRecordingActivated = false;
// Pause loopback if it should only be active during recording
if (playLoopback && loopbackOnlyWhenActive && _playChannel.hasHandle())
_playChannel.setPaused(true);
// Tear down Whisper stream
if (_streamStarted)
{
try { _stream.StopStream(); } catch { /* ignore */ }
_streamStarted = false;
}
_stream = null;
Debug.Log("[FMOD→Whisper] Stream deactivated (Whisper stopped; FMOD still recording).");
}
private void Update()
{
if (!isRecordingActivated) return;
// Always tick FMOD
if (_core.handle != IntPtr.Zero) _core.update();
if (!_streamStarted || !_recSound.hasHandle()) return;
if (!_recSound.hasHandle()) return;
// How many samples recorded since last frame?
// Compute how many samples recorded since last frame.
uint recPos;
_core.getRecordPosition(recordDriverId, out recPos);
@@ -152,23 +252,39 @@ public class FMODWhisperBridge : MonoBehaviour
? (recPos - _lastRecordPos)
: (recPos + _soundPcmLength - _lastRecordPos);
if (deltaSamples == 0) return;
if (deltaSamples == 0)
{
// Even if 0, keep last pos
_lastRecordPos = recPos;
return;
}
// We’ll read that region (16-bit) and convert to float[] [-1..1].
// Calculate byte range to lock in sound buffer
uint bytesToRead = deltaSamples * (uint)_nativeChannels * 2; // 16-bit = 2 bytes
// If not active, we *still* advance the ring (so we don't backlog data),
// but we *don't* push chunks to Whisper.
bool shouldFeed = isRecordingActivated && _streamStarted && _stream != null;
// Calculate byte range to lock (16-bit)
uint bytesToRead = deltaSamples * (uint)_nativeChannels * 2; // 2 bytes per sample
uint startBytes = _lastRecordPos * (uint)_nativeChannels * 2;
IntPtr p1, p2;
uint len1, len2;
// Lock can wrap — FMOD splits into p1/p2.
_recSound.@lock(startBytes, bytesToRead, out p1, out p2, out len1, out len2);
var r = _recSound.@lock(startBytes, bytesToRead, out p1, out p2, out len1, out len2);
if (r != RESULT.OK)
{
// If lock fails, still advance last position to avoid spin
_lastRecordPos = recPos;
return;
}
try
{
// Convert both parts to float and push to Whisper
if (len1 > 0) CopyPcm16ToFloatAndFeed(p1, len1);
if (len2 > 0) CopyPcm16ToFloatAndFeed(p2, len2);
if (shouldFeed)
{
if (len1 > 0) CopyPcm16ToFloatAndFeed(p1, len1);
if (len2 > 0) CopyPcm16ToFloatAndFeed(p2, len2);
}
// else: just discard; we’re only keeping the ring fresh.
}
finally
{
@@ -178,19 +294,11 @@ public class FMODWhisperBridge : MonoBehaviour
_lastRecordPos = recPos;
}
public void ActivateRecording()
{
isRecordingActivated = true;
}
public void DeactivateRecording()
{
isRecordingActivated = false;
}
private bool IsSpeechMeaningful(string userText)
{
return !string.IsNullOrEmpty(userText) && !userText.Contains("BLANK_AUDIO") && !userText.Trim().Equals("[ Silence ]");
return !string.IsNullOrEmpty(userText)
&& !userText.Contains("BLANK_AUDIO")
&& !userText.Trim().Equals("[ Silence ]");
}
private void CopyPcm16ToFloatAndFeed(IntPtr src, uint byteLen)
@@ -198,26 +306,20 @@ public class FMODWhisperBridge : MonoBehaviour
int samples = (int)(byteLen / 2); // 2 bytes per sample
EnsureTmpCapacity(samples);
// Marshal the 16-bit PCM into managed space
// We pin a short[] overlay to avoid copying twice
int shorts = samples;
int byteCount = (int)byteLen;
EnsureShortOverlay(samples, out short[] sBuf);
Marshal.Copy(src, sBuf, 0, samples);
// Use Marshal.Copy into a short[] then convert to float[-1..1]
// (You can also unsafe copy for speed if needed.)
EnsureShortOverlay(shorts, out short[] sBuf);
Marshal.Copy(src, sBuf, 0, shorts);
for (int i = 0; i < shorts; i++)
// Convert to float [-1..1] (no downmix change from your original)
for (int i = 0; i < samples; i++)
{
// 32768f avoids clipping at -32768
_floatTmp[i] = Mathf.Clamp(sBuf[i] / 32768f, -1f, 1f);
}
// Build a chunk for WhisperStream; with VAD off, IsVoiceDetected=true is fine.
// TODO (optional): downmix to mono and/or run a light gate before feeding.
// For now we keep your original behavior:
var chunk = new AudioChunk
{
Data = _floatTmp.AsSpan(0, shorts).ToArray(),
Data = _floatTmp.AsSpan(0, samples).ToArray(),
Frequency = (_nativeRate > 0) ? _nativeRate : desiredSampleRate,
Channels = _nativeChannels,
IsVoiceDetected = true
@@ -226,7 +328,6 @@ public class FMODWhisperBridge : MonoBehaviour
_stream.AddToStream(chunk);
}
private short[] _shortOverlay;
private void EnsureShortOverlay(int samples, out short[] buf)
{
if (_shortOverlay == null || _shortOverlay.Length < samples)
@@ -242,13 +343,20 @@ public class FMODWhisperBridge : MonoBehaviour
private void OnDisable()
{
if (_streamStarted)
{
_stream.StopStream();
_streamStarted = false;
}
// Stop Whisper (if active)
DeactivateRecording();
if (_playChannel.hasHandle()) { _playChannel.stop(); _playChannel.clearHandle(); }
if (_recSound.hasHandle()) { _core.recordStop(recordDriverId); _recSound.release(); _recSound.clearHandle(); }
// Stop/purge FMOD resources (since object is going away)
if (_playChannel.hasHandle())
{
try { _playChannel.stop(); } catch { /* ignore */ }
_playChannel.clearHandle();
}
if (_recSound.hasHandle())
{
try { _core.recordStop(recordDriverId); } catch { /* ignore */ }
try { _recSound.release(); } catch { /* ignore */ }
_recSound.clearHandle();
}
}
}