1// Copyright 2011 Dmitry Chestnykh. All rights reserved.
  2// Use of this source code is governed by a MIT-style
  3// license that can be found in the LICENSE file.
  4
  5package captcha
  6
  7import (
  8	"bytes"
  9	"encoding/binary"
 10	"io"
 11
 12	"math"
 13	"math/rand"
 14)
 15
 16const sampleRate = 8000 // Hz
 17
 18var (
 19	endingBeepSound []byte
 20)
 21
 22func init() {
 23	endingBeepSound = changeSpeed(beepSound, 1.4)
 24}
 25
 26type Audio struct {
 27	body        *bytes.Buffer
 28	digitSounds [][]byte
 29}
 30
 31// NewImage returns a new audio captcha with the given digits, where each digit
 32// must be in range 0-9. Digits are pronounced in the given language. If there
 33// are no sounds for the given language, English is used.
 34func NewAudio(digits []byte, lang string) *Audio {
 35	a := new(Audio)
 36	if sounds, ok := digitSounds[lang]; ok {
 37		a.digitSounds = sounds
 38	} else {
 39		a.digitSounds = digitSounds["en"]
 40	}
 41	numsnd := make([][]byte, len(digits))
 42	nsdur := 0
 43	for i, n := range digits {
 44		snd := a.randomizedDigitSound(n)
 45		nsdur += len(snd)
 46		numsnd[i] = snd
 47	}
 48	// Random intervals between digits (including beginning).
 49	intervals := make([]int, len(digits)+1)
 50	intdur := 0
 51	for i := range intervals {
 52		dur := rnd(sampleRate, sampleRate*3) // 1 to 3 seconds
 53		intdur += dur
 54		intervals[i] = dur
 55	}
 56	// Generate background sound.
 57	bg := a.makeBackgroundSound(a.longestDigitSndLen()*len(digits) + intdur)
 58	// Create buffer and write audio to it.
 59	sil := makeSilence(sampleRate / 5)
 60	bufcap := 3*len(beepSound) + 2*len(sil) + len(bg) + len(endingBeepSound)
 61	a.body = bytes.NewBuffer(make([]byte, 0, bufcap))
 62	// Write prelude, three beeps.
 63	a.body.Write(beepSound)
 64	a.body.Write(sil)
 65	a.body.Write(beepSound)
 66	a.body.Write(sil)
 67	a.body.Write(beepSound)
 68	// Write digits.
 69	pos := intervals[0]
 70	for i, v := range numsnd {
 71		mixSound(bg[pos:], v)
 72		pos += len(v) + intervals[i+1]
 73	}
 74	a.body.Write(bg)
 75	// Write ending (one beep).
 76	a.body.Write(endingBeepSound)
 77	return a
 78}
 79
 80// WriteTo writes captcha audio in WAVE format into the given io.Writer, and
 81// returns the number of bytes written and an error if any.
 82func (a *Audio) WriteTo(w io.Writer) (n int64, err error) {
 83	// Calculate padded length of PCM chunk data.
 84	bodyLen := uint32(a.body.Len())
 85	paddedBodyLen := bodyLen
 86	if bodyLen%2 != 0 {
 87		paddedBodyLen++
 88	}
 89	totalLen := uint32(len(waveHeader)) - 4 + paddedBodyLen
 90	// Header.
 91	header := make([]byte, len(waveHeader)+4) // includes 4 bytes for chunk size
 92	copy(header, waveHeader)
 93	// Put the length of whole RIFF chunk.
 94	binary.LittleEndian.PutUint32(header[4:], totalLen)
 95	// Put the length of WAVE chunk.
 96	binary.LittleEndian.PutUint32(header[len(waveHeader):], bodyLen)
 97	// Write header.
 98	nn, err := w.Write(header)
 99	n = int64(nn)
100	if err != nil {
101		return
102	}
103	// Write data.
104	n, err = a.body.WriteTo(w)
105	n += int64(nn)
106	if err != nil {
107		return
108	}
109	// Pad byte if chunk length is odd.
110	// (As header has even length, we can check if n is odd, not chunk).
111	if bodyLen != paddedBodyLen {
112		w.Write([]byte{0})
113		n++
114	}
115	return
116}
117
118// EncodedLen returns the length of WAV-encoded audio captcha.
119func (a *Audio) EncodedLen() int {
120	return len(waveHeader) + 4 + a.body.Len()
121}
122
123func (a *Audio) makeBackgroundSound(length int) []byte {
124	b := makeWhiteNoise(length, 4)
125	for i := 0; i < length/(sampleRate/10); i++ {
126		snd := reversedSound(a.digitSounds[rand.Intn(10)])
127		snd = changeSpeed(snd, rndf(0.8, 1.4))
128		place := rand.Intn(len(b) - len(snd))
129		setSoundLevel(snd, rndf(0.2, 0.5))
130		mixSound(b[place:], snd)
131	}
132	return b
133}
134
135func (a *Audio) randomizedDigitSound(n byte) []byte {
136	s := randomSpeed(a.digitSounds[n])
137	setSoundLevel(s, rndf(0.75, 1.2))
138	return s
139}
140
141func (a *Audio) longestDigitSndLen() int {
142	n := 0
143	for _, v := range a.digitSounds {
144		if n < len(v) {
145			n = len(v)
146		}
147	}
148	return n
149}
150
151// mixSound mixes src into dst. Dst must have length equal to or greater than
152// src length.
153func mixSound(dst, src []byte) {
154	for i, v := range src {
155		av := int(v)
156		bv := int(dst[i])
157		if av < 128 && bv < 128 {
158			dst[i] = byte(av * bv / 128)
159		} else {
160			dst[i] = byte(2*(av+bv) - av*bv/128 - 256)
161		}
162	}
163}
164
165func setSoundLevel(a []byte, level float64) {
166	for i, v := range a {
167		av := float64(v)
168		switch {
169		case av > 128:
170			if av = (av-128)*level + 128; av < 128 {
171				av = 128
172			}
173		case av < 128:
174			if av = 128 - (128-av)*level; av > 128 {
175				av = 128
176			}
177		default:
178			continue
179		}
180		a[i] = byte(av)
181	}
182}
183
184// changeSpeed returns new PCM bytes from the bytes with the speed and pitch
185// changed to the given value that must be in range [0, x].
186func changeSpeed(a []byte, speed float64) []byte {
187	b := make([]byte, int(math.Floor(float64(len(a))*speed)))
188	var p float64
189	for _, v := range a {
190		for i := int(p); i < int(p+speed); i++ {
191			b[i] = v
192		}
193		p += speed
194	}
195	return b
196}
197
198func randomSpeed(a []byte) []byte {
199	pitch := rndf(0.9, 1.2)
200	return changeSpeed(a, pitch)
201}
202
203func makeSilence(length int) []byte {
204	b := make([]byte, length)
205	for i := range b {
206		b[i] = 128
207	}
208	return b
209}
210
211func makeWhiteNoise(length int, level uint8) []byte {
212	noise := randomBytes(length)
213	adj := 128 - level/2
214	for i, v := range noise {
215		v %= level
216		v += adj
217		noise[i] = v
218	}
219	return noise
220}
221
222func reversedSound(a []byte) []byte {
223	n := len(a)
224	b := make([]byte, n)
225	for i, v := range a {
226		b[n-1-i] = v
227	}
228	return b
229}