Runtime/Inference/Audio/AudioInputService.cs

using System;
using UnityEngine;
namespace OnDeviceAgent.Inference
{

    public sealed class AudioInputService : MonoBehaviour
    {
        public const int TargetSampleRate = 16000;

        [SerializeField] bool m_StartOnEnable = true;
        [SerializeField] int m_MicrophoneLoopSeconds = 10;
        [SerializeField] int m_HistorySeconds = 12;
        [SerializeField] float m_SilenceDetectionSeconds = 4f;
        [SerializeField] float m_SilenceEpsilon = 1e-6f;
        [SerializeField] float m_RestartCooldownSeconds = 5f;

        public event Action<float[], int, long> SamplesReady;
        public bool Mute { get; set; }

        AudioClip m_MicrophoneClip;
        int m_LastMicrophonePosition;
        readonly FloatRingBuffer m_ResampleBuffer = new FloatRingBuffer(TargetSampleRate);
        FloatRingBuffer m_History;
        float[] m_ReadBuffer = Array.Empty<float>();
        float[] m_OutputBuffer = Array.Empty<float>();
        float m_ResampleCursor;
        long m_NextSampleSequence;
        float m_SilentSeconds;
        float m_LastRestartTime = float.MinValue;
        float m_FrameMaxAbs;

        public int HistoryCount => m_History.Count;
        public bool IsRunning => m_MicrophoneClip != null;
        public long NextSampleSequence => m_NextSampleSequence;

        void Awake()
        {
            m_History = new FloatRingBuffer(TargetSampleRate * m_HistorySeconds);
        }

        void OnEnable()
        {
            if (m_StartOnEnable)
                StartMicrophone();
        }

        void Update()
        {
            if (m_MicrophoneClip != null)
                ReadMicrophoneSamples();
        }

        void OnDisable()
        {
            StopMicrophone();
        }

        public void StartMicrophone()
        {
            if (Microphone.devices.Length == 0)
            {
                Debug.LogError("[Voice] No microphone device found.");
                return;
            }

            m_MicrophoneClip = Microphone.Start(null, true, m_MicrophoneLoopSeconds, TargetSampleRate);
            if (m_MicrophoneClip == null)
            {
                Debug.LogError("[Voice] Failed to start default microphone.");
                return;
            }

            m_LastMicrophonePosition = 0;
            m_ResampleCursor = 0f;
            m_ResampleBuffer.Clear();
            m_History.Clear();
            m_NextSampleSequence = 0;

            Debug.Log($"[Voice] Mic started: default device, {m_MicrophoneClip.frequency} Hz, {m_MicrophoneClip.channels} ch");
        }

        public void StopMicrophone()
        {
            if (Microphone.IsRecording(null))
                Microphone.End(null);

            m_MicrophoneClip = null;
            m_LastMicrophonePosition = 0;
        }

        public void CopyLatest(int sampleCount, float[] destination, int destinationIndex)
        {
            m_History.CopyLatest(sampleCount, destination, destinationIndex);
        }

        void ReadMicrophoneSamples()
        {
            if (!Microphone.IsRecording(null))
            {
                TryRestart("[Voice] Mic stopped unexpectedly; restarting");
                return;
            }

            var clipSamples = m_MicrophoneClip.samples;
            var position = Microphone.GetPosition(null);
            if (clipSamples <= 0 || position < 0)
                return;

            if (position >= clipSamples)
                position = 0;

            if (m_LastMicrophonePosition >= clipSamples)
                m_LastMicrophonePosition = 0;

            if (position == m_LastMicrophonePosition)
                return;

            m_FrameMaxAbs = 0f;

            if (position > m_LastMicrophonePosition)
            {
                ReadMicrophoneRange(m_LastMicrophonePosition, position - m_LastMicrophonePosition);
            }
            else
            {
                ReadMicrophoneRange(m_LastMicrophonePosition, clipSamples - m_LastMicrophonePosition);
                ReadMicrophoneRange(0, position);
            }

            m_LastMicrophonePosition = position;

            if (m_FrameMaxAbs >= m_SilenceEpsilon)
            {
                m_SilentSeconds = 0f;
            }
            else
            {
                m_SilentSeconds += Time.unscaledDeltaTime;
                if (m_SilentSeconds >= m_SilenceDetectionSeconds)
                    TryRestart("[Voice] Mic silenced detected; restarting capture");
            }
        }

        void TryRestart(string logMessage)
        {
            if (Time.unscaledTime - m_LastRestartTime < m_RestartCooldownSeconds)
                return;

            Debug.Log(logMessage);
            m_LastRestartTime = Time.unscaledTime;
            m_SilentSeconds = 0f;
            StopMicrophone();
            StartMicrophone();
        }

        void ReadMicrophoneRange(int offsetSamples, int sampleFrames)
        {
            var clipSamples = m_MicrophoneClip.samples;
            if (sampleFrames <= 0 || offsetSamples < 0 || offsetSamples >= clipSamples)
                return;

            sampleFrames = Mathf.Min(sampleFrames, clipSamples - offsetSamples);
            var channels = Mathf.Max(1, m_MicrophoneClip.channels);
            var sampleCount = sampleFrames * channels;
            if (sampleCount <= 0)
                return;

            if (m_ReadBuffer.Length != sampleCount)
                m_ReadBuffer = new float[sampleCount];

            m_MicrophoneClip.GetData(m_ReadBuffer, offsetSamples);

            for (var i = 0; i < sampleCount; i++)
            {
                var abs = m_ReadBuffer[i] < 0f ? -m_ReadBuffer[i] : m_ReadBuffer[i];
                if (abs > m_FrameMaxAbs) m_FrameMaxAbs = abs;
            }

            if (m_MicrophoneClip.frequency == TargetSampleRate && channels == 1)
            {
                PublishSamples(m_ReadBuffer, sampleFrames);
                return;
            }

            var outputCapacity = m_MicrophoneClip.frequency == TargetSampleRate
                ? sampleFrames
                : Mathf.CeilToInt(sampleFrames * (TargetSampleRate / (float)m_MicrophoneClip.frequency)) + 4;
            if (m_OutputBuffer.Length < outputCapacity)
                m_OutputBuffer = new float[outputCapacity];

            var outputCount = m_MicrophoneClip.frequency == TargetSampleRate
                ? DownmixInto(m_ReadBuffer, sampleFrames, channels, m_OutputBuffer)
                : ResampleInto(m_ReadBuffer, sampleFrames, channels, m_MicrophoneClip.frequency, m_OutputBuffer);

            if (outputCount > 0)
                PublishSamples(m_OutputBuffer, outputCount);
        }

        void PublishSamples(float[] samples, int length)
        {
            if (Mute || SupertonicTtsModule.IsSpeaking)
            {
                m_NextSampleSequence += length;
                return;
            }

            var startSequence = m_NextSampleSequence;
            m_NextSampleSequence += length;
            m_History.AddRange(samples, length);
            SamplesReady?.Invoke(samples, length, startSequence);
        }

        static int DownmixInto(float[] source, int sampleFrames, int channels, float[] destination)
        {
            for (var frame = 0; frame < sampleFrames; frame++)
                destination[frame] = Downmix(source, frame, channels);

            return sampleFrames;
        }

        int ResampleInto(float[] source, int sampleFrames, int channels, int sourceRate, float[] destination)
        {
            for (var frame = 0; frame < sampleFrames; frame++)
                m_ResampleBuffer.Add(Downmix(source, frame, channels));

            var outputCount = 0;
            var step = sourceRate / (float)TargetSampleRate;
            while (m_ResampleBuffer.Count >= Mathf.CeilToInt(m_ResampleCursor) + 2 && outputCount < destination.Length)
            {
                var i0 = Mathf.FloorToInt(m_ResampleCursor);
                var t = m_ResampleCursor - i0;
                destination[outputCount++] = Mathf.Lerp(m_ResampleBuffer[i0], m_ResampleBuffer[i0 + 1], t);
                m_ResampleCursor += step;

                var drop = Mathf.FloorToInt(m_ResampleCursor);
                if (drop > 0)
                    m_ResampleBuffer.RemoveFromStart(Mathf.Min(drop, m_ResampleBuffer.Count - 2));
                m_ResampleCursor -= drop;
            }

            return outputCount;
        }

        static float Downmix(float[] data, int frame, int channels)
        {
            var sum = 0f;
            var baseIndex = frame * channels;
            for (var channel = 0; channel < channels; channel++)
                sum += data[baseIndex + channel];

            return sum / channels;
        }
    }
}