/* vplay.c - plays and records CREATIVE LABS VOICE-files, Microsoft WAVE-files and raw data Autor: Michael Beck - beck@informatik.hu-berlin.de */ #include #include #include #include #include #include #include #include #include "fmtheaders.h" #include #define DEFAULT_DSP_SPEED 8000 #define RECORD 0 #define PLAY 1 #define AUDIO "/dev/dsp" #define min(a,b) ((a) <= (b) ? (a) : (b)) #define d_printf(x) if (verbose_mode) fprintf x #define VOC_FMT 0 #define WAVE_FMT 1 #define RAW_DATA 2 #define SND_FMT 3 /* global data */ int timelimit = 0, dsp_speed = DEFAULT_DSP_SPEED, dsp_stereo = 0; int samplesize = 8; int quiet_mode = 0; int verbose_mode = 0; int convert = 0; int record_type = VOC_FMT; u_long count; int audio, abuf_size, zbuf_size; int direction, omode; u_char *audiobuf, *zerobuf; char *command; int vocminor, vocmajor; /* .VOC version */ /* defaults for playing raw data */ struct { int timelimit, dsp_speed, dsp_stereo, samplesize; } raw_info = { 0, DEFAULT_DSP_SPEED, 0, 8}; /* needed prototypes */ void record_play (char *name); void start_voc (int fd, u_long count); void start_wave (int fd, u_long count); void start_snd (int fd, u_long count); void end_voc (int fd); void end_wave_raw (int fd); void end_snd (int fd); u_long calc_count (); struct fmt_record { void (*start) (int fd, u_long count); void (*end) (int fd); char *what; } fmt_rec_table[] = { { start_voc, end_voc, "VOC"} , { start_wave, end_wave_raw, "WAVE"} , { NULL, end_wave_raw, "raw data"} , { start_snd, end_snd, "SND"} }; int main (int argc, char *argv[]) { int c, n = 0; command = argv[0]; if (strstr (argv[0], "vrec")) { direction = RECORD; omode = O_RDONLY; } else if (strstr (argv[0], "vplay")) { direction = PLAY; omode = O_WRONLY; } else { fprintf (stderr, "Error: command should be named either vrec or vplay\n"); exit (1); } while ((c = getopt (argc, argv, "aqs:St:b:vrwd")) != EOF) switch (c) { case 'S': dsp_stereo = raw_info.dsp_stereo = 1; break; case 'q': quiet_mode = 1; break; case 'r': record_type = RAW_DATA; break; case 'v': record_type = VOC_FMT; break; case 'w': record_type = WAVE_FMT; break; case 'a': record_type = SND_FMT; break; case 's': dsp_speed = atoi (optarg); if (dsp_speed < 300) dsp_speed *= 1000; raw_info.dsp_speed = dsp_speed; break; case 't': timelimit = raw_info.timelimit = atoi (optarg); break; case 'b': samplesize = raw_info.samplesize = atoi (optarg); break; case 'd': verbose_mode = 1; quiet_mode = 0; break; default: fprintf (stderr, "Usage: %s [-qvwraS] [-t secs] [-s Hz] [-b 8|12|16] [filename]\n", command); exit (-1); } audio = open (AUDIO, omode, 0); if (audio == -1) { perror (AUDIO); exit (-1); } if (optind > argc - 1) record_play (NULL); else while (optind <= argc - 1) { if (n++ > 0) if (ioctl (audio, SNDCTL_DSP_SYNC, NULL) < 0) { perror (AUDIO); exit (-1); } record_play (argv[optind++]); } close (audio); return 0; } /* test, if it is a .VOC file and return >=0 if ok (this is the length of rest) < 0 if not */ int test_vocfile (void *buffer) { VocHeader *vp = buffer; if (!strcmp ((const char *) vp->magic, MAGIC_STRING)) { vocminor = vp->version & 0xFF; vocmajor = vp->version / 256; if (vp->version != (0x1233 - vp->coded_ver)) return -2; /* coded version mismatch */ return vp->headerlen - sizeof (VocHeader); /* 0 mostly */ } return -1; /* magic string fail */ } /* test, if it's a .WAV file, 0 if ok (and set the speed, stereo etc.) < 0 if not */ int test_wavefile (void *buffer) { WaveHeader *wp = buffer; if (wp->main_chunk == RIFF && wp->chunk_type == WAVE && wp->sub_chunk == FMT && wp->data_chunk == DATA) { if (wp->format != PCM_CODE) { fprintf (stderr, "%s: can't play not PCM-coded WAVE-files\n", command); exit (-1); } if (wp->modus > 2) { fprintf (stderr, "%s: can't play WAVE-files with %d tracks\n", command, wp->modus); exit (-1); } dsp_stereo = (wp->modus == WAVE_STEREO) ? 1 : 0; samplesize = wp->bit_p_spl; dsp_speed = wp->sample_fq; count = wp->data_length; return 0; } return -1; } /* * test, if it's a .SND file, 0 if ok (and set the speed, stereo etc.) * < 0 if not */ int test_sndfile (void *buffer, int fd) { long infolen; char *info; SndHeader *snd = buffer; if (snd->magic == SND_MAGIC) convert = 0; else { if (htonl (snd->magic) == SND_MAGIC) { convert = 1; snd->dataLocation = htonl (snd->dataLocation); snd->dataSize = htonl (snd->dataSize); snd->dataFormat = htonl (snd->dataFormat); snd->samplingRate = htonl (snd->samplingRate); snd->channelCount = htonl (snd->channelCount); } else { /* No SoundFile */ return (-1); } } switch (snd->dataFormat) { case SND_FORMAT_LINEAR_8: samplesize = 8; break; case SND_FORMAT_LINEAR_16: samplesize = 16; break; default: fprintf (stderr, "%s: Unsupported SND_FORMAT\n", command); exit (-1); } dsp_stereo = (snd->channelCount == 2) ? 1 : 0; dsp_speed = snd->samplingRate; count = snd->dataSize; /* read Info-Strings */ infolen = snd->dataLocation - sizeof (SndHeader); info = (char *) malloc (infolen); read (fd, info, infolen); if (!quiet_mode) fprintf (stderr, "SoundFile Info: %s\n", info); free (info); return 0; } /* writing zeros from the zerobuf to simulate silence, perhaps it's enough to use a long var instead of zerobuf ? */ void write_zeros (unsigned x) { unsigned l; while (x) { l = min (x, zbuf_size); if (write (audio, (char *) zerobuf, l) != l) { perror (AUDIO); exit (-1); } x -= l; } } /* if need a SYNC, (is needed if we plan to change speed, stereo ... during output) */ void sync_dsp (void) { #if 0 if (ioctl (audio, SNDCTL_DSP_SYNC, NULL) < 0) { perror (AUDIO); exit (-1); } #endif } /* setting the speed for output */ void set_dsp_speed (int *dsp_speed) { if (ioctl (audio, SNDCTL_DSP_SPEED, dsp_speed) < 0) { fprintf (stderr, "%s: unable to set audio speed\n", command); perror (AUDIO); exit (-1); } } /* if to_mono: compress 8 bit stereo data 2:1, needed if we want play 'one track'; this is usefull, if you habe SB 1.0 - 2.0 (I have 1.0) and want hear the sample (in Mono) if to_8: compress 16 bit to 8 by using hi-byte; wave-files use signed words, so we need to convert it in "unsigned" sample (0x80 is now zero) WARNING: this procedure can't compress 16 bit stereo to 16 bit mono, because if you have a 16 (or 12) bit card you should have stereo (or I'm wrong ?? ) */ u_long one_channel (u_char * buf, u_long l, char to_mono, char to_8) { register u_char *w = buf; register u_char *w2 = buf; char ofs = 0; u_long incr = 0; u_long c, ret; if (to_mono) ++incr; if (to_8) { ++incr; ++w2; ofs = 128; } ret = c = l >> incr; incr = incr << 1; while (c--) { *w++ = *w2 + ofs; w2 += incr; } return ret; } /* ok, let's play a .voc file */ void vplay (int fd, int ofs, char *name) { int l, real_l; BlockType *bp; Voice_data *vd; Ext_Block *eb; u_long nextblock, in_buffer; u_char *data = audiobuf; char was_extended = 0, output = 0; u_short *sp, repeat = 0; u_long silence; int filepos = 0; char one_chn = 0; #define COUNT(x) nextblock -= x; in_buffer -=x ;data += x /* first SYNC the dsp */ sync_dsp (); if (!quiet_mode) fprintf (stderr, "Playing Creative Labs Voice file ...\n"); /* first we waste the rest of header, ugly but we don't need seek */ while (ofs > abuf_size) { read (fd, (char *) audiobuf, abuf_size); ofs -= abuf_size; } if (ofs) read (fd, audiobuf, ofs); /* .voc files are 8 bit (now) */ samplesize = VOC_SAMPLESIZE; ioctl (audio, SNDCTL_DSP_SETFMT, &samplesize); if (samplesize != VOC_SAMPLESIZE) { fprintf (stderr, "%s: unable to set 8 bit sample size!\n", command); exit (-1); } /* and there are MONO by default */ dsp_stereo = MODE_MONO; ioctl (audio, SNDCTL_DSP_STEREO, &dsp_stereo); in_buffer = nextblock = 0; while (1) { Fill_the_buffer: /* need this for repeat */ if (in_buffer < 32) { /* move the rest of buffer to pos 0 and fill the audiobuf up */ if (in_buffer) memcpy ((char *) audiobuf, data, in_buffer); data = audiobuf; if ((l = read (fd, (char *) audiobuf + in_buffer, abuf_size - in_buffer)) > 0) in_buffer += l; else if (!in_buffer) { /* the file is truncated, so simulate 'Terminator' and reduce the datablock for save landing */ nextblock = audiobuf[0] = 0; if (l == -1) { perror (name); exit (-1); } } } while (!nextblock) { /* this is a new block */ bp = (BlockType *) data; COUNT (sizeof (BlockType)); nextblock = DATALEN (bp); if (output && !quiet_mode) fprintf (stderr, "\n"); /* write /n after ASCII-out */ output = 0; switch (bp->type) { case 0: d_printf ((stderr, "Terminator\n")); return; /* VOC-file stop */ case 1: vd = (Voice_data *) data; COUNT (sizeof (Voice_data)); /* we need a SYNC, before we can set new SPEED, STEREO ... */ sync_dsp (); if (!was_extended) { dsp_speed = (int) (vd->tc); dsp_speed = 1000000 / (256 - dsp_speed); d_printf ((stderr, "Voice data %d Hz\n", dsp_speed)); if (vd->pack) { /* /dev/dsp can't it */ fprintf (stderr, "%s: can't play packed .voc files\n", command); return; } if (dsp_stereo) { /* if we are in Stereo-Mode, switch back */ dsp_stereo = MODE_MONO; ioctl (audio, SNDCTL_DSP_STEREO, &dsp_stereo); } } else { /* there was extended block */ if (one_chn) /* if one Stereo fails, why test another ? */ dsp_stereo = MODE_MONO; else if (dsp_stereo) { /* want Stereo */ /* shit, my MACRO dosn't work here */ #ifdef OSS_VERSION if (ioctl (audio, SNDCTL_DSP_STEREO, &dsp_stereo) < 0) { #else if (dsp_stereo != ioctl (audio, SNDCTL_DSP_STEREO, dsp_stereo)) { #endif dsp_stereo = MODE_MONO; fprintf (stderr, "%s: can't play in Stereo; playing only one channel\n", command); one_chn = 1; } } was_extended = 0; } set_dsp_speed (&dsp_speed); break; case 2: /* nothing to do, pure data */ d_printf ((stderr, "Voice continuation\n")); break; case 3: /* a silence block, no data, only a count */ sp = (u_short *) data; COUNT (sizeof (u_short)); dsp_speed = (int) (*data); COUNT (1); dsp_speed = 1000000 / (256 - dsp_speed); sync_dsp (); set_dsp_speed (&dsp_speed); silence = (((u_long) * sp) * 1000) / dsp_speed; d_printf ((stderr, "Silence for %ld ms\n", silence)); write_zeros (*sp); break; case 4: /* a marker for syncronisation, no effect */ sp = (u_short *) data; COUNT (sizeof (u_short)); d_printf ((stderr, "Marker %d\n", *sp)); break; case 5: /* ASCII text, we copy to stderr */ output = 1; d_printf ((stderr, "ASCII - text :\n")); break; case 6: /* repeat marker, says repeatcount */ /* my specs don't say it: maybe this can be recursive, but I don't think somebody use it */ repeat = *(u_short *) data; COUNT (sizeof (u_short)); d_printf ((stderr, "Repeat loop %d times\n", repeat)); if (filepos >= 0) /* if < 0, one seek fails, why test another */ if ((filepos = lseek (fd, 0, 1)) < 0) { fprintf (stderr, "%s: can't play loops; %s isn't seekable\n", command, name); repeat = 0; } else filepos -= in_buffer; /* set filepos after repeat */ else repeat = 0; break; case 7: /* ok, lets repeat that be rewinding tape */ if (repeat) { if (repeat != 0xFFFF) { d_printf ((stderr, "Repeat loop %d\n", repeat)); --repeat; } else d_printf ((stderr, "Neverending loop\n")); lseek (fd, filepos, 0); in_buffer = 0; /* clear the buffer */ goto Fill_the_buffer; } else d_printf ((stderr, "End repeat loop\n")); break; case 8: /* the extension to play Stereo, I have SB 1.0 :-( */ was_extended = 1; eb = (Ext_Block *) data; COUNT (sizeof (Ext_Block)); dsp_speed = (int) (eb->tc); dsp_speed = 256000000L / (65536 - dsp_speed); dsp_stereo = eb->mode; if (dsp_stereo == MODE_STEREO) dsp_speed = dsp_speed >> 1; if (eb->pack) { /* /dev/dsp can't it */ fprintf (stderr, "%s: can't play packed .voc files\n", command); return; } d_printf ((stderr, "Extended block %s %d Hz\n", (eb->mode ? "Stereo" : "Mono"), dsp_speed)); break; default: fprintf (stderr, "%s: unknown blocktype %d. terminate.\n", command, bp->type); return; } /* switch (bp->type) */ } /* while (! nextblock) */ /* put nextblock data bytes to dsp */ l = min (in_buffer, nextblock); if (l) { if (output && !quiet_mode) write (2, data, l); /* to stderr */ else { real_l = one_chn ? one_channel (data, l, one_chn, 0) : l; if (write (audio, data, real_l) != real_l) { perror (AUDIO); exit (-1); } } COUNT (l); } } /* while(1) */ } /* that was a big one, perhaps somebody split it :-) */ /* setting the globals for playing raw data */ void init_raw_data (void) { timelimit = raw_info.timelimit; dsp_speed = raw_info.dsp_speed; dsp_stereo = raw_info.dsp_stereo; samplesize = raw_info.samplesize; } /* calculate the data count to read from/to dsp */ u_long calc_count (void) { u_long count; if (!timelimit) count = 0x7fffffff; else { count = timelimit * dsp_speed; if (dsp_stereo) count *= 2; if (samplesize != 8) count *= 2; } return count; } /* write a .VOC-header */ void start_voc (int fd, u_long cnt) { VocHeader vh; BlockType bt; Voice_data vd; Ext_Block eb; strcpy ((char *) vh.magic, MAGIC_STRING); vh.headerlen = sizeof (VocHeader); vh.version = ACTUAL_VERSION; vh.coded_ver = 0x1233 - ACTUAL_VERSION; write (fd, &vh, sizeof (VocHeader)); if (dsp_stereo) { /* write a extended block */ bt.type = 8; bt.datalen = 4; bt.datalen_m = bt.datalen_h = 0; write (fd, &bt, sizeof (BlockType)); eb.tc = (u_short) (65536 - 256000000L / (dsp_speed << 1)); eb.pack = 0; eb.mode = 1; write (fd, &eb, sizeof (Ext_Block)); } bt.type = 1; cnt += sizeof (Voice_data); /* Voice_data block follows */ bt.datalen = (u_char) (cnt & 0xFF); bt.datalen_m = (u_char) ((cnt & 0xFF00) >> 8); bt.datalen_h = (u_char) ((cnt & 0xFF0000) >> 16); write (fd, &bt, sizeof (BlockType)); vd.tc = (u_char) (256 - (1000000 / dsp_speed)); vd.pack = 0; write (fd, &vd, sizeof (Voice_data)); } /* write a WAVE-header */ void start_wave (int fd, u_long cnt) { WaveHeader wh; wh.main_chunk = RIFF; wh.length = cnt + sizeof (WaveHeader) - 8; wh.chunk_type = WAVE; wh.sub_chunk = FMT; wh.sc_len = 16; wh.format = PCM_CODE; wh.modus = dsp_stereo ? 2 : 1; wh.sample_fq = dsp_speed; wh.byte_p_spl = ((samplesize == 8) ? 1 : 2) * (dsp_stereo ? 2 : 1); wh.byte_p_sec = dsp_speed * wh.modus * wh.byte_p_spl; wh.bit_p_spl = samplesize; wh.data_chunk = DATA; wh.data_length = cnt; write (fd, &wh, sizeof (WaveHeader)); } /* closing .VOC */ void end_voc (int fd) { char dummy = 0; /* Write a Terminator */ write (fd, &dummy, 1); if (fd != 1) close (fd); } void end_wave_raw (int fd) { /* only close output */ if (fd != 1) close (fd); } void start_snd (int fd, u_long count) { SndHeader snd; char *sndinfo = "Recorded by vrec\000"; snd.magic = SND_MAGIC; snd.dataLocation = sizeof (SndHeader) + strlen (sndinfo); snd.dataSize = count; switch (samplesize) { case 8: snd.dataFormat = SND_FORMAT_LINEAR_8; break; case 16: snd.dataFormat = SND_FORMAT_LINEAR_16; break; default: fprintf (stderr, "%d bit: unsupported sample size for NeXt sound file!\n", samplesize); exit (-1); } snd.samplingRate = dsp_speed; snd.channelCount = dsp_stereo ? 2 : 1; write (fd, &snd, sizeof (SndHeader)); write (fd, sndinfo, strlen (sndinfo)); } void end_snd (int fd) { if (fd != 1) close (fd); } /* playing/recording raw data, this proc handels WAVE files and recording .VOCs (as one block) */ void recplay (int fd, int loaded, u_long count, int rtype, char *name) { int l, real_l; u_long c; char one_chn = 0; char to_8 = 0; int tmps; sync_dsp (); tmps = samplesize; ioctl (audio, SNDCTL_DSP_SETFMT, &tmps); if (tmps != samplesize) { fprintf (stderr, "%s: unable to set %d bit sample size", command, samplesize); if (samplesize == 16) { samplesize = 8; ioctl (audio, SNDCTL_DSP_SETFMT, &samplesize); if (samplesize != 8) { fprintf (stderr, "%s: unable to set 8 bit sample size!\n", command); exit (-1); } fprintf (stderr, "; playing 8 bit\n"); to_8 = 1; } else { fprintf (stderr, "\n"); exit (-1); } } #ifdef OSS_VERSION if (ioctl (audio, SNDCTL_DSP_STEREO, &dsp_stereo) < 0) { #else if (dsp_stereo != ioctl (audio, SNDCTL_DSP_STEREO, dsp_stereo)) { #endif if (direction == PLAY) { fprintf (stderr, "%s: can't play in Stereo; playing only one channel\n", command); dsp_stereo = MODE_MONO; one_chn = 1; } else { fprintf (stderr, "%s: can't record in Stereo\n", command); exit (-1); } } set_dsp_speed (&dsp_speed); if (!quiet_mode) { fprintf (stderr, "%s %s : ", (direction == PLAY) ? "Playing" : "Recording", fmt_rec_table[rtype].what); if (samplesize != 8) fprintf (stderr, "%d bit, ", samplesize); fprintf (stderr, "Speed %d Hz ", dsp_speed); fprintf (stderr, "%d bits ", samplesize); fprintf (stderr, "%s ...\n", dsp_stereo ? "Stereo" : "Mono"); } abuf_size = -1; #if 0 /* * There is no idea in using SNDCTL_DSP_GETBLKSIZE in applications like this. * Using some fixed local buffer size will work equally well. */ ioctl (audio, SNDCTL_DSP_GETBLKSIZE, &abuf_size); if (abuf_size == -1) { perror (AUDIO); exit (-1); } #endif zbuf_size = 256; if ((audiobuf = (u_char *) malloc (abuf_size)) == NULL || (zerobuf = (u_char *) malloc (zbuf_size)) == NULL) { fprintf (stderr, "%s: unable to allocate input/output buffer\n", command); exit (-1); } memset ((char *) zerobuf, 128, zbuf_size); if (direction == PLAY) { while (count) { c = count; if (c > abuf_size) c = abuf_size; if ((l = read (fd, (char *) audiobuf + loaded, c - loaded)) > 0) { l += loaded; loaded = 0; /* correct the count; ugly but ... */ real_l = (one_chn || to_8) ? one_channel (audiobuf, l, one_chn, to_8) : l; /* change byte order if necessary */ if (convert && (samplesize == 16)) { long i; for (i = 0; i < real_l; i += 2) *((short *) (audiobuf + i)) = htons (*((short *) (audiobuf + i))); } if (write (audio, (char *) audiobuf, real_l) != real_l) { perror (AUDIO); exit (-1); } count -= l; } else { if (l == -1) { perror (name); exit (-1); } count = 0; /* Stop */ } } /* while (count) */ } else { /* we are recording */ while (count) { c = count; if (c > abuf_size) c = abuf_size; if ((l = read (audio, (char *) audiobuf, c)) > 0) { if (write (fd, (char *) audiobuf, l) != l) { perror (name); exit (-1); } count -= l; } if (l == -1) { perror (AUDIO); exit (-1); } } /* While count */ } } /* let's play or record it (record_type says VOC/WAVE/raw) */ void record_play (char *name) { int fd, ofs; if (direction == PLAY) { if (!name) { fd = 0; name = "stdin"; } else if ((fd = open (name, O_RDONLY, 0)) == -1) { perror (name); exit (-1); } /* Read the smallest header first, then the missing bytes for the next, etc. */ /* read SND-header */ read (fd, (char *) audiobuf, sizeof (SndHeader)); if (test_sndfile (audiobuf, fd) >= 0) recplay (fd, 0, count, SND_FMT, name); else { /* read VOC-Header */ read (fd, (char *) audiobuf + sizeof (SndHeader), sizeof (VocHeader) - sizeof (SndHeader)); if ((ofs = test_vocfile (audiobuf)) >= 0) vplay (fd, ofs, name); else { /* read bytes for WAVE-header */ read (fd, (char *) audiobuf + sizeof (VocHeader), sizeof (WaveHeader) - sizeof (VocHeader)); if (test_wavefile (audiobuf) >= 0) recplay (fd, 0, count, WAVE_FMT, name); else { /* should be raw data */ init_raw_data (); count = calc_count (); recplay (fd, sizeof (WaveHeader), count, RAW_DATA, name); } } } if (fd != 0) close (fd); } else { /* recording ... */ if (!name) { fd = 1; name = "stdout"; } else { if ((fd = open (name, O_WRONLY | O_CREAT | O_TRUNC, 0666)) == -1) { perror (name); exit (-1); } } count = calc_count () & 0xFFFFFFFE; /* WAVE-file should be even (I'm not sure), but wasting one byte isn't a problem (this can only be in 8 bit mono) */ if (fmt_rec_table[record_type].start) fmt_rec_table[record_type].start (fd, count); recplay (fd, 0, count, record_type, name); fmt_rec_table[record_type].end (fd); } }