av/input/audio/audio.go

/*
NAME
  audio.go

AUTHOR
  Alan Noble <alan@ausocean.org>
  Trek Hopton <trek@ausocean.org>

LICENSE
  This file is Copyright (C) 2019 the Australian Ocean Lab (AusOcean)

  It is free software: you can redistribute it and/or modify them
  under the terms of the GNU General Public License as published by the
  Free Software Foundation, either version 3 of the License, or (at your
  option) any later version.

  It is distributed in the hope that it will be useful, but WITHOUT
  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
  for more details.

  You should have received a copy of the GNU General Public License in gpl.txt.
  If not, see [GNU licenses](http://www.gnu.org/licenses).
*/

// Package audio provides access to input from audio devices.
package audio

import (
	"bytes"
	"errors"
	"fmt"
	"sync"
	"time"

	"github.com/yobert/alsa"

	"bitbucket.org/ausocean/av/codec/adpcm"
	"bitbucket.org/ausocean/av/codec/codecutil"
	"bitbucket.org/ausocean/av/codec/pcm"
	"bitbucket.org/ausocean/utils/logger"
	"bitbucket.org/ausocean/utils/ring"
)

const (
	pkg               = "audio: "
	rbTimeout         = 100 * time.Millisecond
	rbNextTimeout     = 100 * time.Millisecond
	rbLen             = 200
	defaultSampleRate = 48000
)

// "running" means the input goroutine is reading from the ALSA device and writing to the ringbuffer.
// "paused" means the input routine is sleeping until unpaused or stopped.
// "stopped" means the input routine is stopped and the ALSA device is closed.
const (
	running = iota + 1
	paused
	stopped
)

// Device holds everything we need to know about the audio input stream and implements io.Reader.
type Device struct {
	l         Logger       // Logger for device's routines to log to.
	mode      uint8        // Operating mode, either running, paused, or stopped.
	mu        sync.Mutex   // Provides synchronisation when changing modes concurrently.
	title     string       // Name of audio title, or empty for the default title.
	dev       *alsa.Device // ALSA's Audio input device.
	ab        alsa.Buffer  // ALSA's buffer.
	rb        *ring.Buffer // Our buffer.
	chunkSize int          // This is the number of bytes that will be stored in rb at a time.
	*Config                // Configuration parameters for this device.
}

// Config provides parameters used by Device.
type Config struct {
	SampleRate int
	Channels   int
	BitDepth   int
	RecPeriod  float64
	Codec      uint8
}

// Logger enables any implementation of a logger to be used.
// TODO: Make this part of the logger package.
type Logger interface {
	SetLevel(int8)
	Log(level int8, message string, params ...interface{})
}

// OpenError is used to determine whether an error has originated from attempting to open a device.
type OpenError error

// NewDevice initializes and returns an Device which can be started, read from, and stopped.
func NewDevice(cfg *Config, l Logger) (*Device, error) {

	err := validate(cfg)
	if err != nil {
		return nil, err
	}

	d := &Device{
		Config: cfg,
		l:      l,
	}

	// Open the requested audio device.
	err = d.open()
	if err != nil {
		d.l.Log(logger.Error, pkg+"failed to open device")
		return nil, err
	}

	// Setup the device to record with desired period.
	d.ab = d.dev.NewBufferDuration(time.Duration(d.RecPeriod * float64(time.Second)))

	// Account for channel conversion.
	chunkSize := float64(len(d.ab.Data) / d.dev.BufferFormat().Channels * d.Channels)

	// Account for resampling.
	chunkSize = (chunkSize / float64(d.dev.BufferFormat().Rate)) * float64(d.SampleRate)
	if chunkSize < 1 {
		return nil, errors.New("given Config parameters are too small")
	}

	// Account for codec conversion.
	if d.Codec == codecutil.ADPCM {
		d.chunkSize = adpcm.EncBytes(int(chunkSize))
	} else {
		d.chunkSize = int(chunkSize)
	}

	// Create ring buffer with appropriate chunk size.
	d.rb = ring.NewBuffer(rbLen, d.chunkSize, rbTimeout)

	// Start device in paused mode.
	d.mode = paused
	go d.input()

	return d, nil
}

// Start will start recording audio and writing to the ringbuffer.
// Once a Device has been stopped it cannot be started again. This is likely to change in future.
func (d *Device) Start() error {
	d.mu.Lock()
	mode := d.mode
	d.mu.Unlock()
	switch mode {
	case paused:
		d.mu.Lock()
		d.mode = running
		d.mu.Unlock()
		return nil
	case stopped:
		// TODO(Trek): Make this reopen device and start recording.
		return errors.New("device is stopped")
	case running:
		return nil
	default:
		return fmt.Errorf("invalid mode: %d", mode)
	}
}

// Stop will stop recording audio and close the device.
// Once a Device has been stopped it cannot be started again. This is likely to change in future.
func (d *Device) Stop() {
	d.mu.Lock()
	d.mode = stopped
	d.mu.Unlock()
}

// ChunkSize returns the number of bytes written to the ringbuffer per d.RecPeriod.
func (d *Device) ChunkSize() int {
	return d.chunkSize
}

// validate checks if Config parameters are valid and returns an error if they are not.
func validate(c *Config) error {
	if c.SampleRate <= 0 {
		return fmt.Errorf("invalid sample rate: %v", c.SampleRate)
	}
	if c.Channels <= 0 {
		return fmt.Errorf("invalid number of channels: %v", c.Channels)
	}
	if c.BitDepth <= 0 {
		return fmt.Errorf("invalid bitdepth: %v", c.BitDepth)
	}
	if c.RecPeriod <= 0 {
		return fmt.Errorf("invalid recording period: %v", c.RecPeriod)
	}
	if !codecutil.IsValid(c.Codec) {
		return errors.New("invalid codec")
	}
	return nil
}

// open the recording device with the given name and prepare it to record.
// If name is empty, the first recording device is used.
func (d *Device) open() error {
	// Close any existing device.
	if d.dev != nil {
		d.l.Log(logger.Debug, pkg+"closing device", "title", d.title)
		d.dev.Close()
		d.dev = nil
	}

	// Open sound card and open recording device.
	d.l.Log(logger.Debug, pkg+"opening sound card")
	cards, err := alsa.OpenCards()
	if err != nil {
		return OpenError(err)
	}
	defer alsa.CloseCards(cards)

	d.l.Log(logger.Debug, pkg+"finding audio device")
	for _, card := range cards {
		devices, err := card.Devices()
		if err != nil {
			continue
		}
		for _, dev := range devices {
			if dev.Type != alsa.PCM || !dev.Record {
				continue
			}
			if dev.Title == d.title || d.title == "" {
				d.dev = dev
				break
			}
		}
	}
	if d.dev == nil {
		return OpenError(errors.New("no audio device found"))
	}

	d.l.Log(logger.Debug, pkg+"opening audio device", "title", d.dev.Title)
	err = d.dev.Open()
	if err != nil {
		return OpenError(err)
	}

	// 2 channels is what most devices need to record in. If mono is requested,
	// the recording will be converted in formatBuffer().
	channels, err := d.dev.NegotiateChannels(2)
	if err != nil {
		return OpenError(err)
	}
	d.l.Log(logger.Debug, pkg+"alsa device channels set", "channels", channels)

	// Try to negotiate a rate to record in that is divisible by the wanted rate
	// so that it can be easily downsampled to the wanted rate.
	// rates is a slice of common sample rates including the standard for CD (44100Hz) and standard for professional audio recording (48000Hz).
	// Note: if a card thinks it can record at a rate but can't actually, this can cause a failure.
	// Eg. the audioinjector sound card is supposed to record at 8000Hz and 16000Hz but it can't due to a firmware issue,
	// a fix for this is to remove 8000 and 16000 from the rates slice.
	var rates = [8]int{8000, 16000, 32000, 44100, 48000, 88200, 96000, 192000}

	var rate int
	foundRate := false
	for r := range rates {
		if r < d.SampleRate {
			continue
		}
		if r%d.SampleRate == 0 {
			rate, err = d.dev.NegotiateRate(r)
			if err == nil {
				foundRate = true
				d.l.Log(logger.Debug, pkg+"alsa device sample rate set", "rate", rate)
				break
			}
		}
	}

	// If no easily divisible rate is found, then use the default rate.
	if !foundRate {
		d.l.Log(logger.Warning, pkg+"Unable to sample at requested rate, default used.", "rateRequested", d.SampleRate)
		rate, err = d.dev.NegotiateRate(defaultSampleRate)
		if err != nil {
			return OpenError(err)
		}
		d.l.Log(logger.Debug, pkg+"alsa device sample rate set", "rate", rate)
	}

	var aFmt alsa.FormatType
	switch d.BitDepth {
	case 16:
		aFmt = alsa.S16_LE
	case 32:
		aFmt = alsa.S32_LE
	default:
		return OpenError(fmt.Errorf("unsupported sample bits %v", d.BitDepth))
	}
	devFmt, err := d.dev.NegotiateFormat(aFmt)
	if err != nil {
		return err
	}
	var bitdepth int
	switch devFmt {
	case alsa.S16_LE:
		bitdepth = 16
	case alsa.S32_LE:
		bitdepth = 32
	default:
		return OpenError(fmt.Errorf("unsupported sample bits %v", d.BitDepth))
	}
	d.l.Log(logger.Debug, pkg+"alsa device bit depth set", "bitdepth", bitdepth)

	// A 50ms period is a sensible value for low-ish latency. (this could be made configurable if needed)
	// Some devices only accept even period sizes while others want powers of 2.
	// So we will find the closest power of 2 to the desired period size.
	const wantPeriod = 0.05 //seconds
	bytesPerSecond := rate * channels * (bitdepth / 8)
	wantPeriodSize := int(float64(bytesPerSecond) * wantPeriod)
	nearWantPeriodSize := nearestPowerOfTwo(wantPeriodSize)

	// At least two period sizes should fit within the buffer.
	bufSize, err := d.dev.NegotiateBufferSize(nearWantPeriodSize * 2)
	if err != nil {
		return OpenError(err)
	}
	d.l.Log(logger.Debug, pkg+"alsa device buffer size set", "buffersize", bufSize)

	if err = d.dev.Prepare(); err != nil {
		return OpenError(err)
	}

	d.l.Log(logger.Debug, pkg+"successfully negotiated ALSA params")
	return nil
}

// input continously records audio and writes it to the ringbuffer.
// Re-opens the device and tries again if ASLA returns an error.
func (d *Device) input() {
	for {
		// Check mode.
		d.mu.Lock()
		mode := d.mode
		d.mu.Unlock()
		switch mode {
		case paused:
			time.Sleep(time.Duration(d.RecPeriod) * time.Second)
			continue
		case stopped:
			if d.dev != nil {
				d.l.Log(logger.Debug, pkg+"closing audio device", "title", d.title)
				d.dev.Close()
				d.dev = nil
			}
			return
		}

		// Read from audio device.
		d.l.Log(logger.Debug, pkg+"recording audio for period", "seconds", d.RecPeriod)
		err := d.dev.Read(d.ab.Data)
		if err != nil {
			d.l.Log(logger.Debug, pkg+"read failed", "error", err.Error())
			err = d.open() // re-open
			if err != nil {
				d.l.Log(logger.Fatal, pkg+"reopening device failed", "error", err.Error())
				return
			}
			continue
		}

		// Process audio.
		d.l.Log(logger.Debug, pkg+"processing audio")
		toWrite := d.formatBuffer()

		// Write audio to ringbuffer.
		n, err := d.rb.Write(toWrite.Data)
		switch err {
		case nil:
			d.l.Log(logger.Debug, pkg+"wrote audio to ringbuffer", "length", n)
		case ring.ErrDropped:
			d.l.Log(logger.Warning, pkg+"old audio data overwritten")
		default:
			d.l.Log(logger.Error, pkg+"unexpected ringbuffer error", "error", err.Error())
			return
		}
	}
}

// Read reads from the ringbuffer, returning the number of bytes read upon success.
func (d *Device) Read(p []byte) (int, error) {
	// Ready ringbuffer for read.
	_, err := d.rb.Next(rbNextTimeout)
	if err != nil {
		return 0, err
	}

	// Read from ring buffer.
	return d.rb.Read(p)
}

// formatBuffer returns audio that has been converted to the desired format.
func (d *Device) formatBuffer() alsa.Buffer {
	var err error

	// If nothing needs to be changed, return the original.
	if d.ab.Format.Channels == d.Channels && d.ab.Format.Rate == d.SampleRate {
		return d.ab
	}
	var formatted alsa.Buffer
	if d.ab.Format.Channels != d.Channels {
		// Convert channels.
		// TODO(Trek): Make this work for conversions other than stereo to mono.
		if d.ab.Format.Channels == 2 && d.Channels == 1 {
			formatted, err = pcm.StereoToMono(d.ab)
			if err != nil {
				d.l.Log(logger.Fatal, pkg+"channel conversion failed", "error", err.Error())
			}
		}
	}

	if d.ab.Format.Rate != d.SampleRate {
		// Convert rate.
		formatted, err = pcm.Resample(formatted, d.SampleRate)
		if err != nil {
			d.l.Log(logger.Fatal, pkg+"rate conversion failed", "error", err.Error())
		}
	}

	switch d.Codec {
	case codecutil.PCM:
	case codecutil.ADPCM:
		b := bytes.NewBuffer(make([]byte, 0, adpcm.EncBytes(len(formatted.Data))))
		enc := adpcm.NewEncoder(b)
		_, err = enc.Write(formatted.Data)
		if err != nil {
			d.l.Log(logger.Fatal, pkg+"unable to encode", "error", err.Error())
		}
		formatted.Data = b.Bytes()
	default:
		d.l.Log(logger.Error, pkg+"unhandled audio codec")
	}

	return formatted
}

// nearestPowerOfTwo finds and returns the nearest power of two to the given integer.
// If the lower and higher power of two are the same distance, it returns the higher power.
// For negative values, 1 is returned.
// Source: https://stackoverflow.com/a/45859570
func nearestPowerOfTwo(n int) int {
	if n <= 0 {
		return 1
	}
	if n == 1 {
		return 2
	}
	v := n
	v--
	v |= v >> 1
	v |= v >> 2
	v |= v >> 4
	v |= v >> 8
	v |= v >> 16
	v++         // higher power of 2
	x := v >> 1 // lower power of 2
	if (v - n) > (n - x) {
		return x
	}
	return v
}