Merge pull request #32 from grahame/timestamp-support

add support for logging timestamps while transcribing
sveinbjornt · Oct 24, 2024 · 176a54a · 176a54a
2 parents 15adb87 + e1d758b
commit 176a54a
Show file tree

Hide file tree

Showing 3 changed files with 35 additions and 4 deletions.
diff --git a/src/Hear.h b/src/Hear.h
@@ -43,6 +43,7 @@
                       onDevice:(BOOL)useOnDeviceRecognition
                 singleLineMode:(BOOL)singleLine
                 addPunctuation:(BOOL)addPunctuation
+                 addTimestamps:(BOOL)addTimestamps
                       exitWord:(NSString *)exitWord
                        timeout:(CGFloat)timeout;
 

diff --git a/src/Hear.m b/src/Hear.m
@@ -49,6 +49,7 @@ @interface Hear()
 @property (nonatomic) BOOL useOnDeviceRecognition;
 @property (nonatomic) BOOL singleLineMode;
 @property (nonatomic) BOOL addPunctuation;
+@property (nonatomic) BOOL addTimestamps;
 @property (nonatomic, retain) NSString *exitWord;
 @property (nonatomic) CGFloat timeout;
 
@@ -61,6 +62,7 @@ - (instancetype)initWithLocale:(NSString *)loc
                       onDevice:(BOOL)onDevice
                 singleLineMode:(BOOL)singleLine
                 addPunctuation:(BOOL)punctuation
+                addTimestamps:(BOOL)timestamps
                       exitWord:(NSString *)exitWord 
                        timeout:(CGFloat)timeout {
     self = [super init];
@@ -76,6 +78,7 @@ - (instancetype)initWithLocale:(NSString *)loc
         self.singleLineMode = singleLine;
         self.useDeviceInput = (input == nil);
         self.addPunctuation = punctuation;
+        self.addTimestamps = timestamps;
         self.exitWord = exitWord;
         self.timeout = timeout;
     }
@@ -193,13 +196,21 @@ - (void)processFile {
         if (result == nil) {
             return;
         }
-
+
+        if (@available(macOS 13, *)) {
+            if (self.addTimestamps) {
+                SFSpeechRecognitionMetadata* meta = result.speechRecognitionMetadata;
+                NSString *timestamp = [[NSDateComponentsFormatter new] stringFromTimeInterval:meta.speechStartTimestamp];
+                NSDump([NSString stringWithFormat:@"\n%@ -> \n", timestamp]);
+            }
+        }
+
         // Make sure there's a space between the incoming result strings
         NSString *s = result.bestTranscription.formattedString;
         if ([s hasSuffix:@" "] == FALSE && !result.isFinal) {
             s = [NSString stringWithFormat:@"%@ ", s];
         }
-        
+
         // Print to stdout without newline and flush
         NSDump(s);
 
@@ -246,7 +257,15 @@ - (void)startListening {
         if (self.timeout > 0) {
             [self startTimer:self];
         }
-
+
+        if (@available(macOS 13, *)) {
+            if (self.addTimestamps) {
+                SFSpeechRecognitionMetadata* meta = result.speechRecognitionMetadata;
+                NSString *timestamp = [[NSDateComponentsFormatter new] stringFromTimeInterval:meta.speechStartTimestamp];
+                NSDump([NSString stringWithFormat:@"\n%@ -> \n", timestamp]);
+            }
+        }
+
         // Print to stdout
         NSString *transcript = result.bestTranscription.formattedString;
         if (self.singleLineMode) {

diff --git a/src/main.m b/src/main.m
@@ -43,7 +43,7 @@
 static inline void PrintHelp(void);
 
 // Command line options
-static const char optstring[] = "sl:i:dpmx:t:hv";
+static const char optstring[] = "sl:i:dpmx:t:Thv";
 
 static struct option long_options[] = {
     // List supported locales for speech to text
@@ -56,6 +56,8 @@
     {"device",                    no_argument,        0, 'd'},
     // Whether to add punctuation to speech recognition results
     {"punctuation",               no_argument,        0, 'p'},
+    // Whether to add timestamps when reading from a file
+    {"timestamp",                 no_argument,        0, 'T'},
     // Enable single-line output mode (for mic)
     {"mode",                      no_argument,        0, 'm'},
     // Exit word
@@ -84,6 +86,7 @@ int main(int argc, const char * argv[]) { @autoreleasepool {
     BOOL useOnDeviceRecognition = NO;
     BOOL singleLineMode = NO;
     BOOL addsPunctuation = NO;
+    BOOL addsTimestamps = NO;
     CGFloat timeout = 0.0f;
 
     // Parse arguments
@@ -124,6 +127,12 @@ int main(int argc, const char * argv[]) { @autoreleasepool {
             case 'p':
                 addsPunctuation = YES;
                 break;
+
+            // Whether to add timestamps to speech recognition results
+            // This option is ignored on macOS versions prior to Ventura
+            case 'T':
+                addsTimestamps = YES;
+                break;
 
             // Set exit word (causes app to exit when word detected in speech)
             case 'x':
@@ -155,6 +164,7 @@ int main(int argc, const char * argv[]) { @autoreleasepool {
                                      onDevice:useOnDeviceRecognition
                                singleLineMode:singleLineMode
                                addPunctuation:addsPunctuation
+                                addTimestamps:addsTimestamps
                                      exitWord:exitWord
                                       timeout:timeout];
     [[NSApplication sharedApplication] setDelegate:hear];
@@ -194,6 +204,7 @@ static inline void PrintHelp(void) {
     -p --punctuation        Add punctuation to speech recognition results (macOS 13+)\n\
     -x --exit-word          Set exit word that causes program to quit\n\
     -t --timeout            Set silence timeout (in seconds)\n\
+    -T --timestamps         Write timestamps as transcription occurs\n\
 \n\
     -h --help               Prints help\n\
     -v --version            Prints program name and version\n\