#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <fcntl.h>
#include <termios.h>
#include <limits.h>
#include <ctype.h>
#include <sys/wait.h>
#include <dirent.h>
#include <errno.h>
#include "HsFFI.h"
#include <espeak-ng/espeak_ng.h>
#include <sndfile.h>
// #define WITH_SPEECHD // FIXME Doesn't support audio cues, navigation, or even read the full page.
#ifdef WITH_SPEECHD
#include <speechd_types.h>
#include <libspeechd.h>
#endif
/* Exported Haskell functions/types */
struct session;
struct session *c_newSession();
void c_freeSession(struct session*);
struct page;
struct page *c_initialReferer();
void *c_fetchURL(struct session*, char*, struct page*, char*);
//struct page **c_fetchURLs(struct session*, struct page*, char**); // FIXME segfaults.
void c_freePage(struct page*);
struct session *c_enableLogging(struct session*);
void c_writeLog(char*, struct session*);
char *c_renderDoc(struct session*, struct page*, _Bool);
char **c_extractLinks(struct page*, char *v2jProfile);
char **c_docLinksAndRendering(struct session*, struct page*, _Bool); // FIXME segfaults.
int c_ssmlHasMark(char*, char*);
char *c_formatLink(char *label, char *title, char *url);
char *c_dading();
char *c_dataDir(char *subdir);
char *c_recognizeIntent(char *profile); // For better JSON & subprocess APIs.
char *c_lastVisited(char*);
/* espeak-ng integration. Based on the espeak-ng command source code. */
SNDFILE *fd_wav = NULL;
char *path_wav = NULL;
static int samplerate;
espeak_ng_ERROR_CONTEXT context;
int choose_format(char *path) {
SF_FORMAT_INFO format_info;
int k, count;
sf_command(fd_wav, SFC_GET_FORMAT_MAJOR_COUNT, &count, sizeof (int));
for (k = 0; k < count; k++) {
format_info.format = k;
sf_command(fd_wav, SFC_GET_FORMAT_MAJOR, &format_info, sizeof (format_info));
char *suffix = path + strlen(path) - strlen(format_info.extension);
if (strcmp(path, format_info.extension) == 0) return format_info.format;
}
return SF_FORMAT_WAV;
}
int paragraph_no = 0, section_no = 0;
int tablerow = 0, tablecol = 0, tableno = 0, in_table = 0;
int capture_marks(short *wav, int numsamples, espeak_EVENT *events) {
while (events->type != 0) {
if (events->type == espeakEVENT_MARK) {
in_table = 0;
if (sscanf(events->id.name, "-rhaps-paragraph%i", ¶graph_no) == 1) {}
else if (sscanf(events->id.name, "-rhaps-section%i", §ion_no) == 1) {}
else if (sscanf(events->id.name, "-rhaps-tablecell%i:%ix%i",
&tableno, &tablerow, &tablecol) == 3) {in_table = 1;}
}
events++;
}
return 0;
}
int save_audio(short *wav, int numsamples, espeak_EVENT *events) {
capture_marks(wav, numsamples, events);
if (wav == NULL) return 0;
while (events->type != 0) {
if (events->type == espeakEVENT_SAMPLERATE) samplerate = events->id.number;
events++;
}
if (fd_wav == NULL) {
SF_INFO info;
info.samplerate = samplerate;
info.channels = 1;
info.format = choose_format(path_wav) | SF_FORMAT_PCM_16 | SF_ENDIAN_LITTLE;
fd_wav = sf_open(path_wav, SFM_WRITE, &info);
}
if (numsamples > 0) sf_writef_short(fd_wav, wav, numsamples);
return 0;
}
int speak_initialize() {
espeak_ng_InitializePath(NULL);
context = NULL;
espeak_ng_STATUS result = espeak_ng_Initialize(&context);
if (result != ENS_OK) {
espeak_ng_PrintStatusCodeMessage(result, stderr, context);
espeak_ng_ClearErrorContext(&context);
return 2;
}
if (path_wav != NULL) {
result = espeak_ng_InitializeOutput(ENOUTPUT_MODE_SYNCHRONOUS, 0, NULL);
espeak_SetSynthCallback(save_audio);
} else {
result = espeak_ng_InitializeOutput(ENOUTPUT_MODE_SPEAK_AUDIO, 0, NULL);
espeak_SetSynthCallback(capture_marks);
}
if (result != ENS_OK) {
espeak_ng_PrintStatusCodeMessage(result, stderr, context);
return 3;
}
samplerate = espeak_ng_GetSampleRate();
return 0;
}
void speak(char *ssml, char *mark, char* fallback) {
int flags = espeakCHARS_AUTO | espeakPHONEMES | espeakENDPAUSE | espeakCHARS_UTF8 | espeakSSML;
if (mark != NULL && c_ssmlHasMark(mark, ssml))
espeak_Synth_Mark(ssml, strlen(ssml)+1, mark, 0, flags, NULL, NULL);
else if (fallback != NULL && c_ssmlHasMark(fallback, ssml))
espeak_Synth_Mark(ssml, strlen(ssml)+1, fallback, 0, flags, NULL, NULL);
else espeak_Synth(ssml, strlen(ssml)+1, 0, POS_CHARACTER, 0, flags, NULL, NULL);
}
void speak_text(char *text, espeak_PARAMETER param, int value) {
int curval = param != 0 ? espeak_GetParameter(param, 0) : 0;
if (param != 0) espeak_SetParameter(param, curval + value, /* relative */0);
int flags = espeakCHARS_AUTO | espeakPHONEMES | espeakENDPAUSE | espeakCHARS_UTF8;
espeak_Synth(text, strlen(text)+1, 0, POS_CHARACTER, 0, flags, NULL, NULL);
if (param != 0) espeak_SetParameter(param, curval, /* relative */0);
}
/* Utilities */
#define SWAP(type, a, b) {type temp = a; a = b; b = temp;}
int min(int a, int b) {return a < b ? a : b;}
/* Keyboard input */
int levenshtein_distance(const char *a, const char* b) {
int a_len = strlen(a);
int b_len = strlen(b);
if (a_len < b_len) {
SWAP(int, a_len, b_len);
SWAP(const char*, a, b);
}
int *v0 = malloc(sizeof(int)*(b_len+1));
int *v1 = malloc(sizeof(int)*(b_len+1));
// Initialize v0 for an edit distance of an empty a
for (int i = 0; i < b_len; i++) v0[i] = i;
// The core algorithm
for (int i = 0; i < a_len; i++) {
// Edit distance from empty string is delete i characters.
v1[0] = i;
for (int j = 0; j < b_len; j++) {
int deletion_cost = v0[j+1] + 1; // above cell + deletion
int insertion_cost = v1[j] + 1; // left cell + insertion
// top left cell + maybe substitution
int substitution_cost = v0[j] + (tolower(a[i]) == tolower(b[j]) ? 0 : 1);
v1[j+1] = min(deletion_cost, min(insertion_cost, substitution_cost));
}
// Progress to next row
SWAP(int*, v0, v1);
}
int ret = v0[b_len];
free(v0); free(v1);
return ret;
}
int MAX_DIST = 3; // Is this ideal?
char *select_link(char **links, const char *command) {
// Pass 1, min distance
int score = INT_MAX;
for (int i = 0; strcmp(links[i], " ") != 0; i++) {
score = min(score, levenshtein_distance(command, links[i]));
}
// Pass 2: Is ambiguous?
int num_matches = 0;
for (int i = 0; strcmp(links[i], " ") != 0; i++)
if (score >= levenshtein_distance(command, links[i]) - MAX_DIST) num_matches++;
espeak_Cancel();
// Pass 3: Retrieve answer
for (int i = 0; strcmp(links[i], " ") != 0; i += 3) {
if (command[0] != '\0' && command[0] != '\n') {// "" to read entire link table...
if (score < levenshtein_distance(command, links[i]) &&
score < levenshtein_distance(command, links[i+1]) &&
score < levenshtein_distance(command, links[i+2])) continue;
if (num_matches == 1) return links[i+2];
}
// Communicate
printf("%s\t%s\t%s\n", links[i+2], links[i], links[i+1]);
char *ssml = c_formatLink(links[i], links[i+1], links[i+2]);
speak(ssml, NULL, NULL);
free(ssml);
}
return NULL;
}
pid_t v2j_waitwake(char *v2j_profile) {
pid_t pid = fork();
if (pid == 0) {/* child */
int pipefds[2];
// Silence standard pipes, not a biggy if it fails.
if (pipe(pipefds) == 0) dup2(pipefds[0], 0);
if (pipe(pipefds) == 0) dup2(pipefds[1], 1);
if (pipe(pipefds) == 0) dup2(pipefds[1], 2);
execlp("voice2json", "voice2json", "--profile", v2j_profile, "wait-wake", "--exit-count", "1", NULL);
}
return pid;
}
volatile pid_t pid_waitwake = 0;
static void check_sigchld(int sig) {
pid_t pid = 0;
do {
pid = waitpid(-1, NULL, WNOHANG);
if (pid == pid_waitwake) pid_waitwake = 0;
} while (pid > 0);
}
struct termios stored_settings, no_echo;
int read_keyboard = 1;
int use_wakeword = 1;
int speak_finalize(char *ssml, char *v2j_profile, char **links, char **out_link) {
// Keyboard input mainloop.
while (read_keyboard) {
if (out_link != NULL && *out_link != NULL) return 0;
char ch = getc(stdin);
if (ch == '\033') {
char mark[200];
char fallback[200];
espeak_Cancel();
char c = getc(stdin);
if (c == 0 || c == -1 || c == '\033' || c == 'q') goto close; // skip [
switch (getc(stdin)) {
case 'A':
// 🠕
if (in_table) {
tablerow--;
if (tablerow > 0) {
sprintf(mark, "-rhaps-tablecell%i:%ix%i", tableno, tablerow, tablecol);
speak(ssml, mark, NULL);
break;
} else in_table = 0;
}
section_no--;
sprintf(mark, "-rhaps-section%i", section_no);
speak(ssml, section_no > 0 ? mark : NULL, NULL);
break;
case 'B':
// 🠗
if (in_table) {
tablerow++;
sprintf(mark, "-rhaps-tablecell%i:%ix%i", tableno, tablerow, tablecol);
sprintf(fallback, "-rhaps-section%i", section_no+1);
speak(ssml, mark, fallback);
break; // FIXME What if that mark doesn't exist?
}
section_no++;
sprintf(mark, "-rhaps-section%i", section_no);
speak(ssml, section_no > 0 ? mark : NULL, NULL);
break;
case 'C':
// ➔
if (in_table) {
tablecol++;
sprintf(mark, "-rhaps-tablecell%i:%ix%i", tableno, tablerow, tablecol);
sprintf(fallback, "-rhaps-paragraph%i", paragraph_no+1);
speak(ssml, mark, fallback);
break; // FIXME What if that mark doesn't exist?
}
paragraph_no++;
sprintf(mark, "-rhaps-paragraph%i", paragraph_no);
speak(ssml, paragraph_no > 0 ? mark : NULL, NULL);
break;
case 'D':
// 🠔
if (in_table) {
tablecol--;
if (tablecol > 0) {
sprintf(mark, "-rhaps-tablecell%i:%ix%i", tableno, tablerow, tablecol);
speak(ssml, mark, NULL);
break;
} else in_table = 0;
}
paragraph_no--;
sprintf(mark, "-rhaps-paragraph%i", paragraph_no);
speak(ssml, paragraph_no > 0 ? mark : NULL, NULL);
break;
}
} else if ((ch == ' ' && v2j_profile != NULL && *v2j_profile != '\0') ||
(use_wakeword && pid_waitwake == 0)) {
espeak_Cancel();
speak(c_dading(), NULL, NULL);
char *line = c_recognizeIntent(v2j_profile);
if (out_link != NULL) {
*out_link = select_link(links, line);
if (*out_link == NULL) pid_waitwake = v2j_waitwake(v2j_profile);
}
free(line);
} else if (ch > 0) {
// Read in a line
tcsetattr(0, TCSANOW, &stored_settings);
char buffer[512];
buffer[0] = ch;
putchar(ch);
char *line = buffer + 1;
size_t len = 510;
if (getline(&line, &len, stdin) < 0)
perror("Failed to read stdin line");
else if (out_link != NULL)
*out_link = select_link(links, buffer);
tcsetattr(0, TCSANOW, &no_echo);
}
}
// Wakeword mainloop
while (use_wakeword && out_link != NULL) {
int status;
pid_t child = v2j_waitwake(v2j_profile);
if (child < 0) continue; // error
if (waitpid(child, &status, 0) == 0) continue;
espeak_Cancel();
char *line = c_recognizeIntent(v2j_profile);
*out_link = select_link(links, line);
if (*out_link != NULL) return 0;
}
// Otherwise, wait until eSpeak has had it's say.
espeak_ng_STATUS result = espeak_ng_Synchronize();
if (result != ENS_OK) {
espeak_ng_PrintStatusCodeMessage(result, stderr, context);
return 4;
}
close:
if (path_wav != NULL) sf_close(fd_wav);
espeak_ng_Terminate();
return 0;
}
/* Main driver */
void write_links(FILE *fd_links, char **links) {
for (int i = 0; strcmp(links[i], " ") != 0; i++) {
fprintf(fd_links, "%s%c", links[i], (i % 3) == 2 ? '\n' : '\t');
}
}
FILE *parse_opt_file() {
FILE *ret = optarg != NULL ? fopen(optarg, "w") : stdout;
if (ret == NULL) {
fprintf(stderr, "Failed to open file %s\n", optarg);
hs_exit();
exit(-1);
}
return ret;
}
int dir_exists(char *path) {
DIR *dir = opendir(path);
if (dir) closedir(dir);
else return ENOENT == errno;
return 0;
}
int main(int argc, char **argv) {
hs_init(&argc, &argv);
int speak_err = 0;
struct session *session = NULL;
struct page *referer = NULL;
int use_espeak = 0;
int validate_v2j_profile = 0;
char *logpath = NULL;
char *mimes;
FILE *fd_ssml = NULL;
FILE *fd_links = NULL;
char *v2j_profile = c_dataDir("voice2json");
tcgetattr(0, &stored_settings);
#ifdef WITH_SPEECHD
SPDConnection *spd_conn = NULL;
#endif
mimes = "text/html text/xml application/xml application/xhtml+xml text/plain";
int c;
opterr = 0;
#ifdef WITH_SPEECHD
while ((c = getopt(argc, argv, "xs::l::L:kKv::VWw::dh")) != -1) {
#else
while ((c = getopt(argc, argv, "xs::l::kKv::VWw::h")) != -1) {
#endif
switch (c) {
case 'x':
mimes = "text/xml application/xml application/xhtml+xml text/html text/plain";
break;
case 's':
fd_ssml = parse_opt_file();
break;
case 'l':
fd_links = parse_opt_file();
break;
case 'L':
logpath = optarg;
break;
case 'K':
read_keyboard = 0;
break;
case 'k':
read_keyboard = 1;
break;
case 'V':
v2j_profile = "";
validate_v2j_profile = 0;
break;
case 'v':
if (optarg != NULL) v2j_profile = optarg;
validate_v2j_profile = 1;
break;
case 'W':
use_wakeword = 0;
break;
case 'w':
use_espeak = 1;
path_wav = optarg;
break;
#ifdef WITH_SPEECHD
case 'd':
spd_conn = spd_open("rhapsode", "main", NULL, SPD_MODE_SINGLE);
if (spd_conn == NULL) fprintf(stderr, "Failed to open SpeechD connection, ignoring\n");
spd_set_data_mode(spd_conn, SPD_DATA_SSML);
break;
#endif
case '?':
fprintf(stderr, "Invalid flag %c\n\n", optopt);
case 'h':
fprintf(stderr, "USAGE: rhapsode [FLAGS] URL...\n");
fprintf(stderr, "\t-x\tX(HT)ML\tIndicates to expect an X(HT)ML file.\n");
fprintf(stderr, "\t-s\tsilent/SSML\tWrites SSML to the specified file or stdout.\n");
fprintf(stderr, "\t\t\thttps://xkcd.com/1692/\n");
fprintf(stderr, "\t-l\tlinks\tWrite extracted links to specifed file or stdout as TSV.\n");
fprintf(stderr, "\t-L\tlog\tWrite (append) network request timing to specified filepath.\n");
fprintf(stderr, "\t-k\tkeyboard in\tRead arrow key navigation & links from stdin. Default behaviour, noop.\n");
fprintf(stderr, "\t-K\t\tDon't read input from stdin.");
fprintf(stderr, "\t-v\tvoice in\tEnsure voice input is enabled & optionally sets the Voice2JSON profile.\n");
fprintf(stderr, "\t-V\t\tDon't listen for voice input.\n");
fprintf(stderr, "\t-W\t\tNo wakeword\tDon't listen for the configured/trained wakeword regardless if voice recognition is enabled.\n");
fprintf(stderr, "\t-w\t.wav\tWrite an audio recording of the webpage, or (DEFAULT) immediately output through speakers.\n");
#ifdef WITH_SPEECHD
fprintf(stderr, "\t-d\tSpeechD\tSchedule page read via the SpeechD daemon. (BROKEN)\n");
#endif
fprintf(stderr, "\t-h\thelp\tOutputs this usage information to stderr.\n");
fprintf(stderr, "\t\t\tIf both -s & -l are enabled without an argument, writes to stderr instead.\n");
hs_exit();
return c == 'h' ? 0 : 1;
}
}
if (read_keyboard) {
// Read input character by character, not line by line.
no_echo = stored_settings;
no_echo.c_lflag &= (~ICANON);
no_echo.c_lflag &= (~ECHO);
no_echo.c_cc[VTIME] = 0;
no_echo.c_cc[VMIN] = 1;
tcsetattr(0, TCSANOW, &no_echo);
}
if (fd_ssml == stdout && fd_links == stdout) fd_links = stderr;
#ifdef WITH_SPEECHD
if (fd_ssml == NULL && fd_links == NULL && spd_conn == NULL && !use_espeak)
#else
if (fd_ssml == NULL && fd_links == NULL && !use_espeak)
#endif
use_espeak = 1;
#if 0 // Unit-test levenshtein_distance
int dist = levenshtein_distance("kitten", "kitten");
printf("kitten => kitten: %i\n", dist);
if (dist != 0) return 10;
dist = levenshtein_distance("kitten", "sitting");
printf("kitten => sitting: %i\n", dist);
if (dist != 3) return 11;
#endif
session = c_newSession();
if (logpath != NULL) session = c_enableLogging(session);
referer = c_initialReferer();
if (use_espeak) speak_err = speak_initialize();
if (validate_v2j_profile && dir_exists(v2j_profile)) {
// TODO Would packagers want a way to configure autodownloading of these profiles?
fprintf(stderr, "Voice2JSON profile is uninitialized! Voice recognition is now disabled.\n");
fprintf(stderr, "Please download a profile from: http://voice2json.org/#supported-languages\n");
fprintf(stderr, "And extract the download to: %s\n", v2j_profile);
if (use_espeak) {
speak_text("Voice2JSON profile is uninitialized! Voice recognition is now disabled.",
espeakRATE, 0);
}
v2j_profile = "";
} else if (dir_exists(v2j_profile)) v2j_profile = "";
use_wakeword &= *v2j_profile != '\0';
char *ssml, **links, *uri;
int read_links = 0;
if (optind >= argc) {
// No URLs specified, restore previous session or goto about:welcome
uri = c_lastVisited("about:welcome");
goto read_uri;
}
for (int i = optind; i < argc; i++) {
uri = argv[i];
read_uri:
if (use_espeak && speak_err == 0) speak_text(uri, espeakRATE, 10);
#ifdef WITH_SPEECHD
else if (spd_conn != NULL) spd_say(spd_conn, SPD_MESSAGE, uri);
#endif
else printf("%s\n", argv[i]);
if (logpath != NULL) session = c_enableLogging(session);
struct page *page = c_fetchURL(session, mimes, referer, uri);
links = c_extractLinks(page, v2j_profile);
ssml = c_renderDoc(session, page, use_espeak);
if (logpath != NULL) c_writeLog(logpath, session);
if (fd_ssml != NULL) fprintf(fd_ssml, "%s\n", ssml);
if (read_keyboard && use_espeak && strcmp(uri, "about:welcome") == 0) read_links = 1;
if (fd_links != NULL) write_links(fd_links, links);
if (use_espeak & speak_err == 0) speak(ssml, "main", NULL);
#ifdef WITH_SPEECHD
if (spd_conn != NULL) spd_say(spd_conn, SPD_MESSAGE, ssml);
#endif
c_freePage(page);
}
uri = NULL;
if (read_links && speak_err == 0) {
speak_err = espeak_ng_Synchronize();
if (speak_err == 0) select_link(links, "");
}
if (read_keyboard && use_wakeword && v2j_profile != NULL && *v2j_profile != 0) {
// Interrupt read when `voice2json wait-wake` exits.
pid_waitwake = v2j_waitwake(v2j_profile);
if (pid_waitwake < 0) {
fprintf(stderr, "Failed to run wakeword detection.\n");
use_wakeword = 0;
}
else {
struct sigaction act;
memset(&act, 0, sizeof(act));
act.sa_handler = check_sigchld;
if (sigaction(SIGCHLD, &act, 0)) {
fprintf(stderr, "Failed to wait upon wakeword detection.\n");
use_wakeword = 0;
}
}
}
if (use_espeak & speak_err == 0) speak_err = speak_finalize(ssml, v2j_profile, links, &uri);
if (uri != NULL) goto read_uri;
#ifdef WITH_SPEECHD
if (spd_conn != NULL) spd_close(spd_conn);
#endif
if (pid_waitwake > 0) {
kill(pid_waitwake, SIGTERM);
kill(pid_waitwake+1, SIGTERM); // Yuck, likely fragile! Also kill voice2json wait-wake's subprocesses.
}
c_freePage(referer);
c_freeSession(session);
hs_exit();
tcsetattr(0, TCSANOW, &stored_settings);
return speak_err;
}