diff --git a/audio/pcm/pcm.go b/audio/pcm/pcm.go new file mode 100644 index 00000000..73342b1b --- /dev/null +++ b/audio/pcm/pcm.go @@ -0,0 +1,124 @@ +package pcm + +import ( + "encoding/binary" + "fmt" + + "github.com/yobert/alsa" +) + +// Resample resamples pcm data (inPcm) from 'fromRate' Hz to 'toRate' Hz and returns the resulting pcm. +// If an error occurs, an error will be returned along with the original audio data +// - channels: number of channels +// - bitDepth: number of bits in single sample +// Notes: +// - Input and output is assumed to be Little Endian. +// - Currently only downsampling is possible and fromRate must be divisible by toRate or an error will occur. +// - If the number of bytes in 'inPcm' is not divisible by the decimation factor (ratioFrom), the remaining bytes will +// not be included in the result. Eg. input of length 480002 downsampling 6:1 will result in output length 80000. +func Resample(inPcm []byte, fromRate, toRate, channels, bitDepth int) ([]byte, error) { + if fromRate == toRate { + return inPcm, nil + } else if fromRate < 0 { + return inPcm, fmt.Errorf("Unable to convert from: %v Hz", fromRate) + } else if toRate < 0 { + return inPcm, fmt.Errorf("Unable to convert to: %v Hz", toRate) + } + + // The number of bytes in a sample. + var sampleLen int + switch bitDepth { + case 32: + sampleLen = 4 * channels + case 16: + sampleLen = 2 * channels + default: + return inPcm, fmt.Errorf("Unhandled bitDepth: %v, must be 16 or 32", bitDepth) + } + inPcmLen := len(inPcm) + + // Calculate sample rate ratio ratioFrom:ratioTo. + rateGcd := gcd(toRate, fromRate) + ratioFrom := fromRate / rateGcd + ratioTo := toRate / rateGcd + + // ratioTo = 1 is the only number that will result in an even sampling. + if ratioTo != 1 { + return inPcm, fmt.Errorf("%v:%v is an unhandled from:to rate ratio. must be n:1 for some rate n", ratioFrom, ratioTo) + } + + newLen := inPcmLen / ratioFrom + result := make([]byte, 0, newLen) + + // For each new sample to be generated, loop through the respective 'ratioFrom' samples in 'inPcm' to add them + // up and average them. The result is the new sample. + for i := 0; i < newLen/sampleLen; i++ { + var sum int + for j := 0; j < ratioFrom; j++ { + switch bitDepth { + case 32: + sum += int(int32(binary.LittleEndian.Uint32(inPcm[(i*ratioFrom*sampleLen)+(j*sampleLen) : (i*ratioFrom*sampleLen)+((j+1)*sampleLen)]))) + case 16: + sum += int(int16(binary.LittleEndian.Uint16(inPcm[(i*ratioFrom*sampleLen)+(j*sampleLen) : (i*ratioFrom*sampleLen)+((j+1)*sampleLen)]))) + default: + return inPcm, fmt.Errorf("Unhandled bitDepth: %v, must be 16 or 32", bitDepth) + } + } + avg := sum / ratioFrom + bAvg := make([]byte, sampleLen) + switch bitDepth { + case 32: + binary.LittleEndian.PutUint32(bAvg, uint32(avg)) + case 16: + binary.LittleEndian.PutUint16(bAvg, uint16(avg)) + } + result = append(result, bAvg...) + } + return result, nil +} + +// StereoToMono returns raw mono audio data generated from only the left channel from +// the given stereo recording (ALSA buffer) +// if an error occurs, an error will be returned along with the original stereo data. +func StereoToMono(stereoBuf alsa.Buffer) ([]byte, error) { + bufChannels := stereoBuf.Format.Channels + if bufChannels == 1 { + return stereoBuf.Data, nil + } else if bufChannels != 2 { + return stereoBuf.Data, fmt.Errorf("Audio is not stereo or mono, it has %v channels", bufChannels) + } + + var stereoSampleBytes int + switch stereoBuf.Format.SampleFormat { + case alsa.S32_LE: + stereoSampleBytes = 8 + case alsa.S16_LE: + stereoSampleBytes = 4 + default: + return stereoBuf.Data, fmt.Errorf("Unhandled ALSA format %v", stereoBuf.Format.SampleFormat) + } + + recLength := len(stereoBuf.Data) + mono := make([]byte, recLength/2) + + // Convert to mono: for each byte in the stereo recording, if it's in the first half of a stereo sample + // (left channel), add it to the new mono audio data. + var inc int + for i := 0; i < recLength; i++ { + if i%stereoSampleBytes < stereoSampleBytes/2 { + mono[inc] = stereoBuf.Data[i] + inc++ + } + } + + return mono, nil +} + +// gcd is used for calculating the greatest common divisor of two positive integers, a and b. +// assumes given a and b are positive. +func gcd(a, b int) int { + if b != 0 { + return gcd(b, a%b) + } + return a +} diff --git a/audio/pcm/pcm_test.go b/audio/pcm/pcm_test.go new file mode 100644 index 00000000..21e66b25 --- /dev/null +++ b/audio/pcm/pcm_test.go @@ -0,0 +1,38 @@ +package pcm + +import ( + "bytes" + "io/ioutil" + "log" + "testing" +) + +// TestResample accepts an input pcm file (assumed to be mono and using 16-bit samples) and outputs a resampled pcm file. +// Input and output file names can be specified as arguments. +func TestResample(t *testing.T) { + inPath := "../../../test/test-data/av/input/sweep_400Hz_20000Hz_-3dBFS_5s_48khz.pcm" + expPath := "../../../test/test-data/av/output/sweep_400Hz_20000Hz_resampled_48to8kHz.pcm" + + // Read input pcm. + inPcm, err := ioutil.ReadFile(inPath) + if err != nil { + log.Fatal(err) + } + + // Resample pcm. + resampled, err := Resample(inPcm, 48000, 8000, 1, 16) + if err != nil { + log.Fatal(err) + } + + // Read expected resampled pcm. + exp, err := ioutil.ReadFile(expPath) + if err != nil { + log.Fatal(err) + } + + // Compare result with expected. + if !bytes.Equal(resampled, exp) { + t.Error("Resampled data does not match expected result.") + } +}