summaryrefslogtreecommitdiffstats
path: root/Whisper/MF/PcmReader.cpp
blob: ab92fc3fe10792a6d3a22041680bf1beeee37d5a (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
#include "stdafx.h"
#include "PcmReader.h"
#include <mfapi.h>
#include "mfUtils.h"

namespace Whisper
{
	__interface iSampleHandler
	{
		void copyChunk( PcmMonoChunk* pMono, const AudioBuffer& rsi, size_t sourceOffset, PcmStereoChunk* pStereo ) const;
		void moveBufferData( AudioBuffer& rdi, size_t amount ) const;
		void appendPcm( AudioBuffer& rdi, const float* rsi, size_t countFloats ) const;
		void copyChunk( PcmMonoChunk* pMono, const AudioBuffer& rsi, size_t sourceOffset, size_t samples, PcmStereoChunk* pStereo ) const;
		uint32_t readerChannelsCount() const;
	};
}

namespace
{
	using namespace Whisper;

	__forceinline void copyMono( PcmMonoChunk* rdi, const AudioBuffer& rsi, size_t sourceOffset, size_t samples )
	{
		assert( sourceOffset + samples <= rsi.mono.size() );
		memcpy( rdi->mono.data(), &rsi.mono[ sourceOffset ], samples * 4 );
		if( samples < FFT_STEP )
			memset( rdi->mono.data() + samples, 0, ( FFT_STEP - samples ) * 4 );
	}

	__forceinline void copyStereo( PcmStereoChunk* rdi, const AudioBuffer& rsi, size_t sourceOffset, size_t samples )
	{
		memcpy( rdi->stereo.data(), &rsi.stereo[ sourceOffset * 2 ], samples * 8 );
		if( samples < FFT_STEP )
			memset( rdi->stereo.data() + samples * 2, 0, ( FFT_STEP - samples ) * 8 );
	}

	struct HandlerMono : iSampleHandler
	{
		void appendPcm( AudioBuffer& rdi, const float* rsi, size_t countFloats ) const override
		{
			rdi.appendMono( rsi, countFloats );
		}
		void copyChunk( PcmMonoChunk* pMono, const AudioBuffer& rsi, size_t sourceOffset, PcmStereoChunk* pStereo ) const override final
		{
			copyMono( pMono, rsi, sourceOffset, FFT_STEP );
		}
		void copyChunk( PcmMonoChunk* pMono, const AudioBuffer& rsi, size_t sourceOffset, size_t samples, PcmStereoChunk* pStereo ) const override final
		{
			copyMono( pMono, rsi, sourceOffset, samples );
		}
		void moveBufferData( AudioBuffer& rdi, size_t amount ) const override final
		{
			const size_t len = rdi.mono.size();
			assert( amount <= len );
			if( amount < len )
			{
				const size_t block = len - amount;
				memmove( rdi.mono.data(), rdi.mono.data() + amount, block * 4 );
				rdi.mono.resize( block );
			}
			else
				rdi.mono.clear();
		}
		uint32_t readerChannelsCount() const override { return 1; }
	};
	struct HandlerDownmixedStereo : HandlerMono
	{
		void appendPcm( AudioBuffer& rdi, const float* rsi, size_t countFloats ) const override final
		{
			rdi.appendDownmixedStereo( rsi, countFloats );
		}
		uint32_t readerChannelsCount() const override final { return 2; }
	};
	struct HandlerStereo : iSampleHandler
	{
		void appendPcm( AudioBuffer& rdi, const float* rsi, size_t countFloats ) const override final
		{
			rdi.appendStereo( rsi, countFloats );
		}
		void copyChunk( PcmMonoChunk* pMono, const AudioBuffer& rsi, size_t sourceOffset, PcmStereoChunk* pStereo ) const override final
		{
			copyMono( pMono, rsi, sourceOffset, FFT_STEP );
			copyStereo( pStereo, rsi, sourceOffset, FFT_STEP );
		}
		void copyChunk( PcmMonoChunk* pMono, const AudioBuffer& rsi, size_t sourceOffset, size_t samples, PcmStereoChunk* pStereo ) const override final
		{
			copyMono( pMono, rsi, sourceOffset, samples );
			copyStereo( pStereo, rsi, sourceOffset, samples );
		}
		void moveBufferData( AudioBuffer& rdi, size_t amount ) const override final
		{
			const size_t len = rdi.mono.size();
			assert( amount <= len );
			if( amount < len )
			{
				const size_t block = len - amount;
				memmove( rdi.mono.data(), rdi.mono.data() + amount, block * 4 );
				rdi.mono.resize( block );
				memmove( rdi.stereo.data(), rdi.stereo.data() + amount * 2, block * 8 );
				rdi.mono.resize( block * 2 );
			}
			else
			{
				rdi.mono.clear();
				rdi.stereo.clear();
			}
		}
		uint32_t readerChannelsCount() const override final { return 2; }
	};
	static const HandlerMono s_mono;
	static const HandlerDownmixedStereo s_downmix;
	static const HandlerStereo s_stereo;
}

PcmReader::PcmReader( IMFSourceReader* reader, bool stereo )
{
	if( nullptr == reader )
		throw E_POINTER;
	this->reader = reader;

	// Set up media type, and figure out sample handler
	check( reader->SetStreamSelection( MF_SOURCE_READER_ALL_STREAMS, FALSE ) );
	check( reader->SetStreamSelection( MF_SOURCE_READER_FIRST_AUDIO_STREAM, TRUE ) );

	CComPtr<IMFMediaType> mtNative;
	check( reader->GetNativeMediaType( MF_SOURCE_READER_FIRST_AUDIO_STREAM, MF_SOURCE_READER_CURRENT_TYPE_INDEX, &mtNative ) );
	UINT32 numChannels;
	check( mtNative->GetUINT32( MF_MT_AUDIO_NUM_CHANNELS, &numChannels ) );

	const bool sourceMono = numChannels < 2;
	if( sourceMono )
		sampleHandler = &s_mono;
	else if( !stereo )
		sampleHandler = &s_downmix;
	else
	{
		sampleHandler = &s_stereo;
		m_stereoOutput = true;
	}

	CComPtr<IMFMediaType> mt;
	check( createMediaType( !sourceMono, &mt ) );
	check( reader->SetCurrentMediaType( MF_SOURCE_READER_FIRST_AUDIO_STREAM, nullptr, mt ) );

	// Find out the length
	int64_t durationTicks;
	check( getStreamDuration( reader, durationTicks ) );

	// Convert length to chunks
	// Seconds = Ticks / 10^7
	// Samples = Seconds * SAMPLE_RATE = Ticks * SAMPLE_RATE / 10^7
	// Chunks = Samples / FFT_STEP = Ticks * SAMPLE_RATE / ( FFT_STEP * 10^7 ), and we want that integer rounded down
	constexpr __int64 mul = SAMPLE_RATE;
	constexpr __int64 div = (__int64)FFT_STEP * 10'000'000;
	m_length = (size_t)MFllMulDiv( durationTicks, mul, div, 0 );
}

HRESULT PcmReader::readNextSample()
{
	const size_t off = bufferReadOffset;
	const size_t availableSamples = pcm.mono.size() - off;

	// If needed, move the remaining PCM data to the start of these vectors
	if( availableSamples > 0 )
	{
		if( 0 != off )
			sampleHandler->moveBufferData( pcm, off );
	}
	else
		pcm.clear();
	bufferReadOffset = 0;

	while( true )
	{
		DWORD dwFlags = 0;
		CComPtr<IMFSample> sample;

		// Read the next sample
		HRESULT hr = reader->ReadSample( (DWORD)MF_SOURCE_READER_FIRST_AUDIO_STREAM, 0, nullptr, &dwFlags, nullptr, &sample );
		if( FAILED( hr ) )
		{
			logErrorHr( hr, u8"IMFSourceReader.ReadSample" );
			return hr;
		}

		if( dwFlags & MF_SOURCE_READERF_CURRENTMEDIATYPECHANGED )
		{
			// logError( u8"Media type changes ain’t supported by the library." );
			// return E_UNEXPECTED;

			// This happens for some video files at the very start of the reading, with Dolby AC3 audio track.
			// Instead of failing the transcribe process, verify the important attributes (FP32 samples, sample rate, count of channels) haven’t changed.
			CHECK( validateCurrentMediaType( reader, sampleHandler->readerChannelsCount() ) );
		}

		if( dwFlags & MF_SOURCE_READERF_ENDOFSTREAM )
			return E_EOF;

		if( !sample )
		{
			// printf( "No sample\n" );
			continue;
		}

		// Get a pointer to the audio data in the sample.
		CComPtr<IMFMediaBuffer> buffer;
		hr = sample->ConvertToContiguousBuffer( &buffer );
		if( FAILED( hr ) )
			return hr;

		const float* pAudioData = nullptr;
		DWORD cbBuffer;
		hr = buffer->Lock( (BYTE**)&pAudioData, nullptr, &cbBuffer );
		if( FAILED( hr ) )
			return hr;

		try
		{
			assert( 0 == ( cbBuffer % sizeof( float ) ) );
			const size_t countFloats = cbBuffer / sizeof( float );
			sampleHandler->appendPcm( pcm, pAudioData, countFloats );
		}
		catch( const std::bad_alloc& )
		{
			buffer->Unlock();
			return E_OUTOFMEMORY;
		}

		// Unlock the buffer
		hr = buffer->Unlock();
		if( FAILED( hr ) )
			return hr;

		return S_OK;
	}
}

HRESULT PcmReader::readChunk( PcmMonoChunk& mono, PcmStereoChunk* stereo )
{
	while( true )
	{
		const size_t off = bufferReadOffset;
		const size_t availableSamples = pcm.mono.size() - off;
		if( availableSamples >= FFT_STEP )
		{
			// We have enough data in the buffer
			sampleHandler->copyChunk( &mono, pcm, off, stereo );
			bufferReadOffset = off + FFT_STEP;
			return S_OK;
		}

		if( !m_readerEndOfFile )
		{
			// We don't have enough data, but the stream has not ended yet, can load moar samples from the reader
			HRESULT hr = readNextSample();
			if( SUCCEEDED( hr ) )
				continue;
			if( hr != E_EOF )
				return hr;
			m_readerEndOfFile = true;
		}

		if( availableSamples > 0 )
		{
			// We have reached the end of stream of the reader, but the buffer still has a few samples.
			// Return the final incomplete chunk padded with zeros
			sampleHandler->copyChunk( &mono, pcm, off, availableSamples, stereo );
			bufferReadOffset = off + availableSamples;
			return S_OK;
		}

		return E_EOF;
	}
}