Refactored audio streams

* All streams are now mono (simplifies reasoning about samples)
* Streams can be cloned
* Streams can be seeked within

CMakeLists.txt

@@ -100,7 +100,7 @@ set(SOURCE_FILES
 	src/phoneExtraction.cpp
 	src/platformTools.cpp
 	src/tools.cpp
-	src/audioInput/ChannelDownmixer.cpp
+	src/audioInput/AudioStream.cpp
 	src/audioInput/SampleRateConverter.cpp
 	src/audioInput/WaveFileReader.cpp
 	src/audioInput/waveFileWriting.cpp

src/audioInput/AudioStream.cpp

@@ -0,0 +1,5 @@
+#include "AudioStream.h"
+
+bool AudioStream::endOfStream() {
+	return getSampleIndex() >= getSampleCount();
+}

src/audioInput/AudioStream.h

@@ -1,9 +1,16 @@
 #pragma once
 
+#include <memory>
+
+// A mono stream of floating-point samples.
 class AudioStream {
 public:
-	virtual int getFrameRate() = 0;
-	virtual int getFrameCount() = 0;
-	virtual int getChannelCount() = 0;
-	virtual bool getNextSample(float &sample) = 0;
+	virtual ~AudioStream() {}
+	virtual std::unique_ptr<AudioStream> clone(bool reset) = 0;
+	virtual int getSampleRate() = 0;
+	virtual int getSampleCount() = 0;
+	virtual int getSampleIndex() = 0;
+	virtual void seek(int sampleIndex) = 0;
+	bool endOfStream();
+	virtual float readSample() = 0;
 };

src/audioInput/ChannelDownmixer.cpp

@@ -1,31 +0,0 @@
-#include "ChannelDownmixer.h"
-
-ChannelDownmixer::ChannelDownmixer(std::unique_ptr<AudioStream> inputStream) :
-	inputStream(std::move(inputStream)),
-	inputChannelCount(this->inputStream->getChannelCount())
-{}
-
-int ChannelDownmixer::getFrameRate() {
-	return inputStream->getFrameRate();
-}
-
-int ChannelDownmixer::getFrameCount() {
-	return inputStream->getFrameCount();
-}
-
-int ChannelDownmixer::getChannelCount() {
-	return 1;
-}
-
-bool ChannelDownmixer::getNextSample(float &sample) {
-	float sum = 0;
-	for (int channelIndex = 0; channelIndex < inputChannelCount; channelIndex++) {
-		float currentSample;
-		if (!inputStream->getNextSample(currentSample)) return false;
-
-		sum += currentSample;
-	}
-
-	sample = sum / inputChannelCount;
-	return true;
-}

src/audioInput/ChannelDownmixer.h

@@ -1,18 +0,0 @@
-#pragma once
-
-#include "AudioStream.h"
-#include <memory>
-
-// Converts a multi-channel audio stream to mono.
-class ChannelDownmixer : public AudioStream {
-public:
-	ChannelDownmixer(std::unique_ptr<AudioStream> inputStream);
-	virtual int getFrameRate() override;
-	virtual int getFrameCount() override;
-	virtual int getChannelCount() override;
-	virtual bool getNextSample(float &sample) override;
-
-private:
-	std::unique_ptr<AudioStream> inputStream;
-	int inputChannelCount;
-};

src/audioInput/SampleRateConverter.cpp

@@ -1,80 +1,94 @@
 #include <cmath>
 #include "SampleRateConverter.h"
+#include <stdexcept>
+#include <algorithm>
 
 using std::runtime_error;
 
 SampleRateConverter::SampleRateConverter(std::unique_ptr<AudioStream> inputStream, int outputFrameRate) :
 	inputStream(std::move(inputStream)),
-	downscalingFactor(static_cast<double>(this->inputStream->getFrameRate()) / outputFrameRate),
+	downscalingFactor(static_cast<double>(this->inputStream->getSampleRate()) / outputFrameRate),
 	outputFrameRate(outputFrameRate),
-	outputFrameCount(std::lround(this->inputStream->getFrameCount() / downscalingFactor)),
+	outputFrameCount(std::lround(this->inputStream->getSampleCount() / downscalingFactor)),
 	lastInputSample(0),
 	lastInputSampleIndex(-1),
 	nextOutputSampleIndex(0)
 {
-	if (this->inputStream->getChannelCount() != 1) {
-		throw runtime_error("Only mono input streams are supported.");
-	}
-	if (this->inputStream->getFrameRate() < outputFrameRate) {
+	if (this->inputStream->getSampleRate() < outputFrameRate) {
 		throw runtime_error("Upsampling not supported.");
 	}
 }
 
-int SampleRateConverter::getFrameRate() {
+SampleRateConverter::SampleRateConverter(const SampleRateConverter& rhs, bool reset) :
+	SampleRateConverter(rhs.inputStream->clone(reset), outputFrameRate)
+{
+	nextOutputSampleIndex = reset ? 0 : rhs.nextOutputSampleIndex;
+}
+
+std::unique_ptr<AudioStream> SampleRateConverter::clone(bool reset) {
+	return std::make_unique<SampleRateConverter>(*this, reset);
+}
+
+int SampleRateConverter::getSampleRate() {
 	return outputFrameRate;
 }
 
-int SampleRateConverter::getFrameCount() {
+int SampleRateConverter::getSampleCount() {
 	return outputFrameCount;
 }
 
-int SampleRateConverter::getChannelCount() {
-	return 1;
+int SampleRateConverter::getSampleIndex() {
+	return nextOutputSampleIndex;
 }
 
-bool SampleRateConverter::getNextSample(float &sample) {
-	if (nextOutputSampleIndex >= outputFrameCount) return false;
+void SampleRateConverter::seek(int sampleIndex) {
+	if (sampleIndex < 0 || sampleIndex >= outputFrameCount) throw std::invalid_argument("sampleIndex out of range.");
 
-	double start = nextOutputSampleIndex * downscalingFactor;
-	double end = (nextOutputSampleIndex + 1) * downscalingFactor;
+	nextOutputSampleIndex = sampleIndex;
+}
+
+float SampleRateConverter::readSample() {
+	if (nextOutputSampleIndex >= outputFrameCount) throw std::out_of_range("End of stream.");
+
+	double inputStart = nextOutputSampleIndex * downscalingFactor;
+	double inputEnd = (nextOutputSampleIndex + 1) * downscalingFactor;
 
-	sample = mean(start, end);
 	nextOutputSampleIndex++;
-	return true;
+	return mean(inputStart, inputEnd);
 }
 
-float SampleRateConverter::mean(double start, double end) {
+float SampleRateConverter::mean(double inputStart, double inputEnd) {
 	// Calculate weighted sum...
 	double sum = 0;
 
 	// ... first sample (weight <= 1)
-	int startIndex = static_cast<int>(start);
-	sum += getInputSample(startIndex) * ((startIndex + 1) - start);
+	int startIndex = static_cast<int>(inputStart);
+	sum += getInputSample(startIndex) * ((startIndex + 1) - inputStart);
 
 	// ... middle samples (weight 1 each)
-	int endIndex = static_cast<int>(end);
+	int endIndex = static_cast<int>(inputEnd);
 	for (int index = startIndex + 1; index < endIndex; index++) {
 		sum += getInputSample(index);
 	}
 
 	// ... last sample (weight < 1)
-	sum += getInputSample(endIndex) * (end - endIndex);
+	sum += getInputSample(endIndex) * (inputEnd - endIndex);
 
-	return static_cast<float>(sum / (end - start));
+	return static_cast<float>(sum / (inputEnd - inputStart));
 }
 
 float SampleRateConverter::getInputSample(int sampleIndex) {
+	sampleIndex = std::min(sampleIndex, inputStream->getSampleCount() - 1);
+	if (sampleIndex < 0) return 0.0f;
+
 	if (sampleIndex == lastInputSampleIndex) {
 		return lastInputSample;
 	}
-	if (sampleIndex == lastInputSampleIndex + 1) {
-		lastInputSampleIndex++;
-		// Read the next sample.
-		// If the input stream has no more samples (at the very end),
-		// we'll just reuse the last sample we have.
-		inputStream->getNextSample(lastInputSample);
+
+	if (sampleIndex != inputStream->getSampleIndex()) {
+		inputStream->seek(sampleIndex);
+	}
+	lastInputSample = inputStream->readSample();
+	lastInputSampleIndex = sampleIndex;
 	return lastInputSample;
 }
-
-	throw runtime_error("Can only return the last sample or the one following it.");
-}

src/audioInput/SampleRateConverter.h

@@ -1,22 +1,21 @@
 #pragma once
 
 #include <memory>
-#include <vector>
 #include "AudioStream.h"
 
 class SampleRateConverter : public AudioStream {
 public:
 	SampleRateConverter(std::unique_ptr<AudioStream> inputStream, int outputFrameRate);
-	virtual int getFrameRate() override;
-	virtual int getFrameCount() override;
-	virtual int getChannelCount() override;
-	virtual bool getNextSample(float &sample) override;
+	SampleRateConverter(const SampleRateConverter& rhs, bool reset);
+	std::unique_ptr<AudioStream> clone(bool reset) override;
+	int getSampleRate() override;
+	int getSampleCount() override;
+	int getSampleIndex() override;
+	void seek(int sampleIndex) override;
+	float readSample() override;
 private:
-	// The stream we're reading from
 	std::unique_ptr<AudioStream> inputStream;
-
-	// input frame rate / output frame rate
-	double downscalingFactor;
+	double downscalingFactor;					// input frame rate / output frame rate
 
 	int outputFrameRate;
 	int outputFrameCount;

src/audioInput/WaveFileReader.cpp

@@ -24,10 +24,12 @@ enum class Codec {
 	Float = 0x03
 };
 
-WaveFileReader::WaveFileReader(boost::filesystem::path filePath) {
-	// Open file
-	file.exceptions(std::ifstream::failbit | std::ifstream::badbit);
-	file.open(filePath, std::ios::binary);
+WaveFileReader::WaveFileReader(boost::filesystem::path filePath) :
+	filePath(filePath),
+	file(),
+	sampleIndex(0)
+{
+	openFile();
 
 	// Read header
 	uint32_t rootChunkId = read<uint32_t>(file);
@@ -42,7 +44,7 @@ WaveFileReader::WaveFileReader(boost::filesystem::path filePath) {
 
 	// Read chunks until we reach the data chunk
 	bool reachedDataChunk = false;
-	int bytesPerSample = 0;
+	bytesPerSample = 0;
 	do {
 		uint32_t chunkId = read<uint32_t>(file);
 		int chunkSize = read<uint32_t>(file);
@@ -56,7 +58,7 @@ WaveFileReader::WaveFileReader(boost::filesystem::path filePath) {
 			int frameSize = read<uint16_t>(file);
 			int bitsPerSample = read<uint16_t>(file);
 
-				// We're read 16 bytes so far. Skip the remainder.
+			// We've read 16 bytes so far. Skip the remainder.
 			file.seekg(roundToEven(chunkSize) - 16, file.cur);
 
 			// Determine sample format
@@ -97,8 +99,9 @@ WaveFileReader::WaveFileReader(boost::filesystem::path filePath) {
 		}
 		case fourcc('d', 'a', 't', 'a'): {
 			reachedDataChunk = true;
-				remainingSamples = chunkSize / bytesPerSample;
-				frameCount = remainingSamples / channelCount;
+			dataOffset = file.tellg();
+			sampleCount = chunkSize / bytesPerSample;
+			frameCount = sampleCount / channelCount;
 			break;
 		}
 		default: {
@@ -110,43 +113,79 @@ WaveFileReader::WaveFileReader(boost::filesystem::path filePath) {
 	} while (!reachedDataChunk);
 }
 
-int WaveFileReader::getFrameRate() {
+WaveFileReader::WaveFileReader(const WaveFileReader& rhs, bool reset) :
+	filePath(rhs.filePath),
+	file(),
+	bytesPerSample(rhs.bytesPerSample),
+	sampleFormat(rhs.sampleFormat),
+	frameRate(rhs.frameRate),
+	frameCount(rhs.frameCount),
+	channelCount(rhs.channelCount),
+	sampleCount(rhs.sampleCount),
+	dataOffset(rhs.dataOffset),
+	sampleIndex(-1)
+{
+	openFile();
+	seek(reset ? 0 : rhs.sampleIndex);
+}
+
+std::unique_ptr<AudioStream> WaveFileReader::clone(bool reset) {
+	return std::make_unique<WaveFileReader>(*this, reset);
+}
+
+void WaveFileReader::openFile() {
+	file.exceptions(std::ifstream::failbit | std::ifstream::badbit);
+	file.open(filePath, std::ios::binary);
+}
+
+int WaveFileReader::getSampleRate() {
 	return frameRate;
 }
 
-int WaveFileReader::getFrameCount() {
+int WaveFileReader::getSampleCount() {
 	return frameCount;
 }
 
-int WaveFileReader::getChannelCount() {
-	return channelCount;
+int WaveFileReader::getSampleIndex() {
+	return sampleIndex;
 }
 
-bool WaveFileReader::getNextSample(float &sample) {
-	if (remainingSamples == 0) return false;
-	remainingSamples--;
+void WaveFileReader::seek(int sampleIndex) {
+	if (sampleIndex < 0 || sampleIndex >= sampleCount) throw std::invalid_argument("sampleIndex out of range.");
 
+	file.seekg(dataOffset + sampleIndex * channelCount * bytesPerSample);
+	this->sampleIndex = sampleIndex;
+}
+
+float WaveFileReader::readSample() {
+	if (sampleIndex + channelCount > sampleCount) throw std::out_of_range("End of stream.");
+	sampleIndex += channelCount;
+
+	float sum = 0;
+	for (int channelIndex = 0; channelIndex < channelCount; channelIndex++) {
 		switch (sampleFormat) {
 		case SampleFormat::UInt8: {
 			uint8_t raw = read<uint8_t>(file);
-			sample = toNormalizedFloat(raw, 0, UINT8_MAX);
+			sum += toNormalizedFloat(raw, 0, UINT8_MAX);
 			break;
 		}
 		case SampleFormat::Int16: {
 			int16_t raw = read<int16_t>(file);
-			sample = toNormalizedFloat(raw, INT16_MIN, INT16_MAX);
+			sum += toNormalizedFloat(raw, INT16_MIN, INT16_MAX);
 			break;
 		}
 		case SampleFormat::Int24: {
 			int raw = read<int, 24>(file);
 			if (raw & 0x800000) raw |= 0xFF000000; // Fix two's complement
-			sample = toNormalizedFloat(raw, INT24_MIN, INT24_MAX);
+			sum += toNormalizedFloat(raw, INT24_MIN, INT24_MAX);
 			break;
 		}
 		case SampleFormat::Float32: {
-			sample = read<float>(file);
+			sum += read<float>(file);
 			break;
 		}
 		}
-	return true;
+	}
+
+	return sum / channelCount;
 }

src/audioInput/WaveFileReader.h

@@ -1,8 +1,5 @@
 #pragma once
 
-#include <string>
-#include <cstdint>
-#include <fstream>
 #include <boost/filesystem/path.hpp>
 #include <boost/filesystem/fstream.hpp>
 #include "AudioStream.h"
@@ -17,16 +14,26 @@ enum class SampleFormat {
 class WaveFileReader : public AudioStream {
 public:
 	WaveFileReader(boost::filesystem::path filePath);
-	virtual int getFrameRate() override ;
-	virtual int getFrameCount() override;
-	virtual int getChannelCount() override;
-	virtual bool getNextSample(float &sample) override;
+	WaveFileReader(const WaveFileReader& rhs, bool reset);
+	std::unique_ptr<AudioStream> clone(bool reset) override;
+	int getSampleRate() override ;
+	int getSampleCount() override;
+	int getSampleIndex() override;
+	void seek(int sampleIndex) override;
+	float readSample() override;
 
 private:
+	void openFile();
+
+private:
+	boost::filesystem::path filePath;
 	boost::filesystem::ifstream file;
+	int bytesPerSample;
 	SampleFormat sampleFormat;
 	int frameRate;
 	int frameCount;
 	int channelCount;
-	int remainingSamples;
+	int sampleCount;
+	size_t dataOffset;
+	int sampleIndex;
 };

src/audioInput/waveFileWriting.cpp

@@ -13,9 +13,9 @@ void createWaveFile(std::unique_ptr<AudioStream> inputStream, std::string fileNa
 	// Write RIFF chunk
 	write<uint32_t>(fourcc('R', 'I', 'F', 'F'), file);
 	uint32_t formatChunkSize = 16;
-	uint16_t channelCount = static_cast<uint16_t>(inputStream->getChannelCount());
+	uint16_t channelCount = 1;
 	uint16_t frameSize = static_cast<uint16_t>(channelCount * sizeof(float));
-	uint32_t dataChunkSize = static_cast<uint32_t>(inputStream->getFrameCount() * frameSize);
+	uint32_t dataChunkSize = static_cast<uint32_t>(inputStream->getSampleCount() * frameSize);
 	uint32_t riffChunkSize = 4 + (8 + formatChunkSize) + (8 + dataChunkSize);
 	write<uint32_t>(riffChunkSize, file);
 	write<uint32_t>(fourcc('W', 'A', 'V', 'E'), file);
@@ -26,7 +26,7 @@ void createWaveFile(std::unique_ptr<AudioStream> inputStream, std::string fileNa
 	uint16_t codec = 0x03; // 32-bit float
 	write<uint16_t>(codec, file);
 	write<uint16_t>(channelCount, file);
-	uint32_t frameRate = static_cast<uint16_t>(inputStream->getFrameRate());
+	uint32_t frameRate = static_cast<uint16_t>(inputStream->getSampleRate());
 	write<uint32_t>(frameRate, file);
 	uint32_t bytesPerSecond = frameRate * frameSize;
 	write<uint32_t>(bytesPerSecond, file);
@@ -37,8 +37,8 @@ void createWaveFile(std::unique_ptr<AudioStream> inputStream, std::string fileNa
 	// Write data chunk
 	write<uint32_t>(fourcc('d', 'a', 't', 'a'), file);
 	write<uint32_t>(dataChunkSize, file);
-	float sample;
-	while (inputStream->getNextSample(sample)) {
+	while (!inputStream->endOfStream()) {
+		float sample = inputStream->readSample();
 		write<float>(sample, file);
 	}
 }

src/main.cpp

@@ -107,7 +107,7 @@ int main(int argc, char *argv[]) {
 		{
 			ProgressBar progressBar;
 			phones = detectPhones(
-				[&inputFileName]() { return createAudioStream(inputFileName.getValue()); },
+				createAudioStream(inputFileName.getValue()),
 				dialog.getValue(),
 				progressBar);
 		}

src/phoneExtraction.cpp

@@ -3,7 +3,6 @@
 #include <boost/algorithm/string.hpp>
 #include "phoneExtraction.h"
 #include "audioInput/SampleRateConverter.h"
-#include "audioInput/ChannelDownmixer.h"
 #include "platformTools.h"
 #include "tools.h"
 #include <format.h>
@@ -33,17 +32,12 @@ using std::regex;
 using std::regex_replace;
 using std::chrono::duration;
 
-unique_ptr<AudioStream> to16kHzMono(unique_ptr<AudioStream> stream) {
-	// Downmix, if required
-	if (stream->getChannelCount() != 1) {
-		stream.reset(new ChannelDownmixer(std::move(stream)));
-	}
-
+unique_ptr<AudioStream> to16kHz(unique_ptr<AudioStream> stream) {
 	// Downsample, if required
-	if (stream->getFrameRate() < 16000) {
+	if (stream->getSampleRate() < 16000) {
 		throw invalid_argument("Audio sample rate must not be below 16kHz.");
 	}
-	if (stream->getFrameRate() != 16000) {
+	if (stream->getSampleRate() != 16000) {
 		stream.reset(new SampleRateConverter(std::move(stream), 16000));
 	}
 
@@ -88,7 +82,7 @@ int16_t floatSampleToInt16(float sample) {
 	return static_cast<int16_t>(((sample + 1) / 2) * (INT16_MAX - INT16_MIN) + INT16_MIN);
 }
 
-void processAudioStream(AudioStream& audioStream16kHzMono, function<void(const vector<int16_t>&)> processBuffer, ProgressSink& progressSink) {
+void processAudioStream(AudioStream& audioStream16kHz, function<void(const vector<int16_t>&)> processBuffer, ProgressSink& progressSink) {
 	// Process entire sound file
 	vector<int16_t> buffer;
 	const int capacity = 1600; // 0.1 second capacity
@@ -97,10 +91,9 @@ void processAudioStream(AudioStream& audioStream16kHzMono, function<void(const v
 	do {
 		// Read to buffer
 		buffer.clear();
-		while (buffer.size() < capacity) {
+		while (buffer.size() < capacity && !audioStream16kHz.endOfStream()) {
 			// Read sample
-			float floatSample;
-			if (!audioStream16kHzMono.getNextSample(floatSample)) break;
+			float floatSample = audioStream16kHz.readSample();
 			int16_t sample = floatSampleToInt16(floatSample);
 			buffer.push_back(sample);
 		}
@@ -109,7 +102,7 @@ void processAudioStream(AudioStream& audioStream16kHzMono, function<void(const v
 		processBuffer(buffer);
 
 		sampleCount += buffer.size();
-		progressSink.reportProgress(static_cast<double>(sampleCount) / audioStream16kHzMono.getFrameCount());
+		progressSink.reportProgress(static_cast<double>(sampleCount) / audioStream16kHz.getSampleCount());
 	} while (buffer.size());
 }
 
@@ -158,7 +151,7 @@ void sphinxLogCallback(void* user_data, err_lvl_t errorLevel, const char* format
 
 vector<string> recognizeWords(unique_ptr<AudioStream> audioStream, ps_decoder_t& recognizer, ProgressSink& progressSink) {
 	// Convert audio stream to the exact format PocketSphinx requires
-	audioStream = to16kHzMono(std::move(audioStream));
+	audioStream = to16kHz(std::move(audioStream));
 
 	// Start recognition
 	int error = ps_start_utt(&recognizer);
@@ -243,7 +236,7 @@ map<centiseconds, Phone> getPhoneAlignment(const vector<s3wid_t>& wordIds, uniqu
 	if (error) throw runtime_error("Error populating alignment struct.");
 
 	// Convert audio stream to the exact format PocketSphinx requires
-	audioStream = to16kHzMono(std::move(audioStream));
+	audioStream = to16kHz(std::move(audioStream));
 
 	// Create search structure
 	acmod_t* acousticModel = recognizer.acmod;
@@ -304,7 +297,7 @@ map<centiseconds, Phone> getPhoneAlignment(const vector<s3wid_t>& wordIds, uniqu
 }
 
 map<centiseconds, Phone> detectPhones(
-	std::function<std::unique_ptr<AudioStream>(void)> createAudioStream,
+	unique_ptr<AudioStream> audioStream,
 	boost::optional<std::string> dialog,
 	ProgressSink& progressSink)
 {
@@ -329,13 +322,13 @@ map<centiseconds, Phone> detectPhones(
 		// Get words
 		vector<string> words = dialog
 			? extractDialogWords(*dialog)
-			: recognizeWords(createAudioStream(), *recognizer.get(), wordRecognitionProgressSink);
+			: recognizeWords(audioStream->clone(true), *recognizer.get(), wordRecognitionProgressSink);
 
 		// Look up words in dictionary
 		vector<s3wid_t> wordIds = getWordIds(words, *recognizer->dict);
 
 		// Align the word's phones with speech
-		map<centiseconds, Phone> result = getPhoneAlignment(wordIds, createAudioStream(), *recognizer.get(), alignmentProgressSink);
+		map<centiseconds, Phone> result = getPhoneAlignment(wordIds, std::move(audioStream), *recognizer.get(), alignmentProgressSink);
 		return result;
 	}
 	catch (...) {

src/phoneExtraction.h

@@ -2,7 +2,6 @@
 
 #include <map>
 #include <memory>
-#include <functional>
 #include "audioInput/AudioStream.h"
 #include "Phone.h"
 #include "centiseconds.h"
@@ -10,6 +9,6 @@
 #include <boost/optional/optional.hpp>
 
 std::map<centiseconds, Phone> detectPhones(
-	std::function<std::unique_ptr<AudioStream>(void)> createAudioStream,
+	std::unique_ptr<AudioStream> audioStream,
 	boost::optional<std::string> dialog,
 	ProgressSink& progressSink);