add silence threshold for great good

src/main.zig

@@ -26,31 +26,36 @@ const SpeechHandler = struct {
     const max_children = 5;
     const Process = struct { child: ?*std.process.Child, start: i64, id: std.process.Child.Id }; // why id?
     /// Handle detected speech
-    fn onSpeech(ctx: *anyopaque, text: []const u8) void {
+    fn onSpeech(ctx: *anyopaque, event: stt.SpeechEvent) void {
         if (builtin.is_test) return; // Suppress output during tests
         // Look for noise words and skip it if so
-        if (std.mem.eql(u8, text, "huh")) return;
-        if (std.mem.eql(u8, text, "but")) return;
+        if (std.mem.eql(u8, event.text, "huh")) return;
+        if (std.mem.eql(u8, event.text, "but")) return;
         const self: *SpeechHandler = @ptrCast(@alignCast(ctx));
         self.speech_count += 1;
 
-        // Print with timestamp for better experience
-        const timestamp = std.time.timestamp();
+        // Print with amplitude for better experience
         var stdout_buffer: [1024]u8 = undefined;
         var stdout_writer = std.fs.File.stdout().writer(&stdout_buffer);
         const stdout = &stdout_writer.interface;
         defer stdout.flush() catch std.log.warn("Caught error writing speech data to stdout", .{});
-        stdout.print("[{}] Speech {}->{?s}: {s}\n", .{
-            timestamp,
+        stdout.print("[{}-{}] Speech {}->{?s}: {s}\n", .{
+            event.min_amplitude,
+            event.max_amplitude,
             self.speech_count,
             self.exec_program,
-            text,
+            event.text,
         }) catch std.log.warn("Caught error writing speech data to stdout", .{});
 
         // Execute program if specified
-        if (self.exec_program) |program| self.exec(text) catch |err| {
-            std.log.err("Failed to execute program '{s}': {}", .{ program, err });
-        };
+        if (self.exec_program) |program| {
+            self.exec(event.text) catch |err| {
+                std.log.err("Failed to execute program '{s}': {}", .{ program, err });
+            };
+            self.reclaimProcessesPosix(false) catch |err| {
+                std.log.err("Failed to reclaim processes: {}", .{err});
+            };
+        }
     }
     fn exec(self: *SpeechHandler, text: []const u8) !void {
         const program = self.exec_program.?; // should only be called when exec_program is not null
@@ -78,7 +83,6 @@ const SpeechHandler = struct {
         try process.child.?.spawn();
         try process.child.?.waitForSpawn();
         process.id = process.child.?.id;
-        try self.reclaimProcessesPosix(false);
     }
 
     fn reclaimProcessesPosix(self: *SpeechHandler, reap_all: bool) !void {
@@ -282,6 +286,70 @@ fn signalHandler(sig: i32) callconv(.c) void {
     }
 }
 
+fn runMeasureLevels(allocator: std.mem.Allocator) !void {
+    const stdout = std.fs.File.stdout();
+    const is_tty = stdout.isTty();
+
+    var capture = try stt.AlsaCapture.init(allocator, "default", 16000, 1024);
+    defer capture.deinit();
+    try capture.open();
+
+    _ = try stdout.writeAll("Measuring audio levels... Press Ctrl+C to exit\n");
+    if (is_tty) {
+        _ = try stdout.writeAll("Histogram (0-10000):\n");
+    }
+
+    var buffer: [4096]i16 = undefined;
+    var write_buffer: [256]u8 = undefined;
+    var second_max: u16 = 0;
+    var last_print = std.time.milliTimestamp();
+
+    while (!should_exit.load(.acquire)) {
+        _ = try capture.readAudio();
+        const samples_read = capture.getAudioSamples(&buffer);
+        if (samples_read == 0) {
+            std.Thread.sleep(10 * std.time.ns_per_ms);
+            continue;
+        }
+
+        var max_amp: u16 = 0;
+        for (buffer[0..samples_read]) |sample| {
+            const abs_sample = @abs(sample);
+            if (abs_sample > max_amp) max_amp = abs_sample;
+        }
+
+        if (max_amp > second_max) second_max = max_amp;
+
+        const now = std.time.milliTimestamp();
+        if (now - last_print >= 1000) {
+            if (is_tty) {
+                const bar_width = (@as(u32, second_max) * 60) / 10000;
+                var writer = stdout.writer(&write_buffer);
+                const w = &writer.interface;
+                try w.print("{d:5} |", .{second_max});
+                try w.flush();
+                for (0..bar_width) |_| {
+                    _ = try stdout.writeAll("█");
+                }
+                _ = try stdout.writeAll("\n");
+            } else {
+                var writer = stdout.writer(&write_buffer);
+                const w = &writer.interface;
+                try w.print("{d}\n", .{second_max});
+                try w.flush();
+            }
+            second_max = 0;
+            last_print = now;
+        }
+
+        std.Thread.sleep(50 * std.time.ns_per_ms);
+    }
+
+    if (is_tty) {
+        _ = try stdout.writeAll("\n");
+    }
+}
+
 fn signalAction(sig: i32, info: *const std.posix.siginfo_t, _: ?*anyopaque) callconv(.c) void {
     // NOTE: info only works correctly if std.posix.SA.SIGINFO is in the flags
     // std.log.debug("signal action. sig {d}", .{sig});
@@ -349,6 +417,8 @@ pub fn main() !void {
 
     var model_path: ?[]const u8 = null;
     var exec_program: ?[]const u8 = null;
+    var measure_levels = false;
+    var silence_threshold: ?i16 = null;
 
     // Parse arguments
     for (args[1..]) |arg| {
@@ -357,13 +427,16 @@ pub fn main() !void {
             _ = try stdout.writeAll("USAGE:\n");
             _ = try stdout.writeAll("    stt [OPTIONS]\n\n");
             _ = try stdout.writeAll("OPTIONS:\n");
-            _ = try stdout.writeAll("    --model=<path>    Path to Vosk model directory\n");
-            _ = try stdout.writeAll("    --exec=<program>  Program to execute with recognized text\n");
-            _ = try stdout.writeAll("    --help, -h        Show this help message\n\n");
+            _ = try stdout.writeAll("    --model=<path>            Path to Vosk model directory\n");
+            _ = try stdout.writeAll("    --exec=<program>          Program to execute with recognized text\n");
+            _ = try stdout.writeAll("    --silence-threshold=<n>   Silence detection threshold (default: 300)\n");
+            _ = try stdout.writeAll("    --measure-levels          Display real-time audio level histogram\n");
+            _ = try stdout.writeAll("    --help, -h                Show this help message\n\n");
             _ = try stdout.writeAll("EXAMPLES:\n");
             _ = try stdout.writeAll("    stt\n");
             _ = try stdout.writeAll("    stt --model=../share/vosk/models/vosk-model-small-en-us-0.15\n");
-            _ = try stdout.writeAll("    stt --exec=echo\n\n");
+            _ = try stdout.writeAll("    stt --exec=echo\n");
+            _ = try stdout.writeAll("    stt --measure-levels\n\n");
             _ = try stdout.writeAll("The application will search for models in these locations:\n");
             _ = try stdout.writeAll("    vosk-model-small-en-us-0.15\n");
             _ = try stdout.writeAll("    <binary_dir>/../share/vosk/models/vosk-model-small-en-us-0.15\n");
@@ -372,6 +445,18 @@ pub fn main() !void {
             model_path = arg[8..]; // Skip "--model="
         } else if (std.mem.startsWith(u8, arg, "--exec=")) {
             exec_program = arg[7..]; // Skip "--exec="
+        } else if (std.mem.eql(u8, arg, "--measure-levels")) {
+            measure_levels = true;
+        } else if (std.mem.startsWith(u8, arg, "--silence-threshold=")) {
+            const threshold_str = arg[20..]; // Skip "--silence-threshold="
+            silence_threshold = std.fmt.parseInt(i16, threshold_str, 10) catch {
+                std.log.err("Invalid silence threshold: {s}", .{threshold_str});
+                return error.InvalidArgument;
+            };
+            if (silence_threshold.? < 0) {
+                std.log.err("Invalid silence threshold: {s}", .{threshold_str});
+                return error.InvalidArgument;
+            }
         }
     }
 
@@ -381,6 +466,13 @@ pub fn main() !void {
         .exec_program = exec_program,
     };
     defer handler.deinit();
+
+    // If measure-levels mode, run that instead of normal STT
+    if (measure_levels) {
+        try runMeasureLevels(allocator);
+        return;
+    }
+
     const speech_handler = stt.SpeechEventHandler{
         .onSpeechFn = SpeechHandler.onSpeech,
         .onErrorFn = SpeechHandler.onError,
@@ -436,6 +528,7 @@ pub fn main() !void {
         .event_handler = speech_handler,
         .sample_rate = 16000, // Standard sample rate for speech recognition
         .buffer_size = 256, // Existing buffer size for low latency
+        .silence_threshold = silence_threshold orelse 300,
     };
 
     std.log.debug("Initializing STT library...", .{});
@@ -526,7 +619,8 @@ test "handler callbacks" {
     };
 
     // Test that callbacks can be invoked without crashing
-    speech_handler.onSpeech("test speech");
+    const event = stt.SpeechEvent{ .text = "test speech", .max_amplitude = 500, .min_amplitude = 200 };
+    speech_handler.onSpeech(event);
     speech_handler.onError(stt.Error.AudioDeviceError, "test error");
 
     // If we get here without crashing, the test passes

src/stt.zig

@@ -104,12 +104,24 @@ pub const ErrorInfo = struct {
     }
 };
 
+/// Speech detection event data
+pub const SpeechEvent = struct {
+    /// Recognized text
+    text: []const u8,
+    /// Maximum amplitude detected in the speech segment
+    max_amplitude: u16 = std.math.maxInt(u16),
+    /// Minimum chunk amplitude that triggered speech detection
+    min_amplitude: u16 = 0,
+    /// Audio data for the speech segment
+    audio_data: []const i16 = &[_]i16{},
+};
+
 /// Callback function type for speech detection events
 ///
 /// Parameters:
-/// - text: Null-terminated string containing the detected speech
+/// - event: Speech event data containing text and amplitude
 /// - user_data: Optional user-provided context data
-pub const SpeechCallback = *const fn (text: [*:0]const u8, user_data: ?*anyopaque) void;
+pub const SpeechCallback = *const fn (event: SpeechEvent, user_data: ?*anyopaque) void;
 
 /// Callback function type for error events
 ///
@@ -132,7 +144,7 @@ pub const DetailedErrorCallback = *const fn (error_info: ErrorInfo, user_data: ?
 /// with both speech detection and error handling callbacks.
 pub const SpeechEventHandler = struct {
     /// Function to call when speech is detected
-    onSpeechFn: *const fn (ctx: *anyopaque, text: []const u8) void,
+    onSpeechFn: *const fn (ctx: *anyopaque, event: SpeechEvent) void,
     /// Function to call when an error occurs
     onErrorFn: *const fn (ctx: *anyopaque, error_code: Error, message: []const u8) void,
     /// Optional function to call for detailed error information
@@ -141,11 +153,11 @@ pub const SpeechEventHandler = struct {
     ctx: *anyopaque,
 
     /// Invoke the speech detection callback with error handling
-    pub fn onSpeech(self: SpeechEventHandler, text: []const u8) void {
+    pub fn onSpeech(self: SpeechEventHandler, event: SpeechEvent) void {
         // Call the speech callback function
         // Note: If the callback panics or causes undefined behavior,
         // there's not much we can do to recover gracefully in Zig
-        self.onSpeechFn(self.ctx, text);
+        self.onSpeechFn(self.ctx, event);
     }
 
     /// Invoke the error callback
@@ -519,7 +531,7 @@ pub const AlsaCapture = struct {
     }
 
     /// Read audio data from ALSA device and process it
-    fn readAudio(self: *Self) !usize {
+    pub fn readAudio(self: *Self) !usize {
         if (self.pcm_handle == null)
             return Error.AudioDeviceError;
 
@@ -594,6 +606,8 @@ pub const Options = struct {
     // channels: u32 = 2,
     /// Audio buffer size in frames (default: 256)
     buffer_size: u32 = 256,
+    /// Silence detection threshold (default: 300)
+    silence_threshold: i16 = 300,
 };
 
 /// Main STT session handle
@@ -958,11 +972,15 @@ pub const Session = struct {
         defer self.allocator.free(vosk_buffer);
 
         // Silence detection parameters
-        const silence_threshold: i16 = 500;
+        const silence_threshold: i16 = self.options.silence_threshold;
         const silence_duration_ms: u64 = 500;
         const min_speech_duration_ms: u64 = 300;
         const samples_per_ms = 16; // This assumes 16kHz audio
 
+        std.log.info(
+            "Listening with silence threshold {}. If wrong, use --measure-levels to find approriate val",
+            .{silence_threshold},
+        );
         var speech_buffer = self.allocator.alloc(i16, 16000 * 10) catch {
             const error_info = ErrorInfo.init(Error.OutOfMemory, "Failed to allocate speech buffer");
             self.options.event_handler.onDetailedError(error_info);
@@ -974,6 +992,8 @@ pub const Session = struct {
         var silence_samples: usize = 0;
         var speech_samples: usize = 0;
         var speech_pos: usize = 0;
+        var speech_max_amplitude: u16 = 0;
+        var speech_min_amplitude: u16 = std.math.maxInt(u16);
 
         while (!self.should_stop.load(.acquire)) {
             const available = self.vosk_audio_buffer.available();
@@ -1002,7 +1022,11 @@ pub const Session = struct {
                     speech_samples = 0;
                     silence_samples = 0;
                     speech_pos = 0;
+                    speech_max_amplitude = 0;
+                    speech_min_amplitude = std.math.maxInt(u16);
                 }
+                if (max_amplitude > speech_max_amplitude) speech_max_amplitude = max_amplitude;
+                if (max_amplitude < speech_min_amplitude) speech_min_amplitude = max_amplitude;
                 const copy_len = @min(samples_read, speech_buffer.len - speech_pos);
                 @memcpy(speech_buffer[speech_pos .. speech_pos + copy_len], vosk_buffer[0..copy_len]);
                 speech_pos += copy_len;
@@ -1016,7 +1040,14 @@ pub const Session = struct {
 
                 if (silence_samples >= silence_duration_ms * samples_per_ms) {
                     if (speech_samples >= min_speech_duration_ms * samples_per_ms) {
-                        self.processVoskAudio(speech_buffer[0..speech_pos]) catch |err| {
+                        const event = SpeechEvent{
+                            // SAFETY: This will be defined in the next line when we process the audio data
+                            .text = undefined,
+                            .max_amplitude = speech_max_amplitude,
+                            .min_amplitude = speech_min_amplitude,
+                            .audio_data = speech_buffer[0..speech_pos],
+                        };
+                        self.processVoskAudio(event) catch |err| {
                             std.log.err("Error processing speech: {}", .{err});
                         };
                     }
@@ -1025,6 +1056,8 @@ pub const Session = struct {
                     speech_pos = 0;
                     speech_samples = 0;
                     silence_samples = 0;
+                    speech_max_amplitude = 0;
+                    speech_min_amplitude = std.math.maxInt(u16);
 
                     if (self.vosk_recognizer) |rec| {
                         c.vosk_recognizer_reset(rec);
@@ -1038,24 +1071,24 @@ pub const Session = struct {
     }
 
     /// Process complete audio clip with Vosk and handle results
-    fn processVoskAudio(self: *Session, audio_data: []const i16) !void {
+    fn processVoskAudio(self: *Session, event: SpeechEvent) !void {
         if (self.vosk_recognizer == null) return Error.InvalidState;
 
-        const audio_bytes = std.mem.sliceAsBytes(audio_data);
+        const audio_bytes = std.mem.sliceAsBytes(event.audio_data);
         _ = c.vosk_recognizer_accept_waveform(self.vosk_recognizer, audio_bytes.ptr, @intCast(audio_bytes.len));
 
         // Get final result
         const result_cstr = c.vosk_recognizer_final_result(self.vosk_recognizer);
         if (result_cstr != null) {
             const result_str = std.mem.span(result_cstr);
-            self.parseVoskResult(result_str) catch |err| {
+            self.parseVoskResult(result_str, event) catch |err| {
                 self.options.event_handler.onError(err, "Failed to parse Vosk result");
             };
         }
     }
 
     /// Parse Vosk JSON result and extract recognized text
-    fn parseVoskResult(self: *Session, json_str: []const u8) !void {
+    fn parseVoskResult(self: *Session, json_str: []const u8, event: SpeechEvent) !void {
         if (json_str.len == 0) return;
 
         const text_key = "\"text\"";
@@ -1074,7 +1107,9 @@ pub const Session = struct {
                     if (std.mem.indexOf(u8, json_str[s..], "\"")) |quote_end| {
                         const text = json_str[s .. s + quote_end];
                         if (text.len > 0 and !std.mem.eql(u8, text, " ")) {
-                            self.options.event_handler.onSpeech(text);
+                            var result_event = event;
+                            result_event.text = text;
+                            self.options.event_handler.onSpeech(result_event);
                         }
                     }
                 }
@@ -1376,9 +1411,9 @@ test "Options validation" {
 
     // Test valid options
     const DummyHandler = struct {
-        fn onSpeech(ctx: *anyopaque, text: []const u8) void {
+        fn onSpeech(ctx: *anyopaque, event: SpeechEvent) void {
             _ = ctx;
-            _ = text;
+            _ = event;
         }
         fn onError(ctx: *anyopaque, error_code: Error, message: []const u8) void {
             _ = ctx;
@@ -1412,9 +1447,9 @@ test "Session state management" {
     const testing = std.testing;
 
     const DummyHandler = struct {
-        fn onSpeech(ctx: *anyopaque, text: []const u8) void {
+        fn onSpeech(ctx: *anyopaque, event: SpeechEvent) void {
             _ = ctx;
-            _ = text;
+            _ = event;
         }
         fn onError(ctx: *anyopaque, error_code: Error, message: []const u8) void {
             _ = ctx;
@@ -1451,12 +1486,14 @@ test "SpeechEventHandler interface" {
         speech_called: bool = false,
         error_called: bool = false,
         last_text: []const u8 = "",
+        last_amplitude: u16 = 0,
         last_error: Error = Error.InitializationFailed,
 
-        fn onSpeech(ctx: *anyopaque, text: []const u8) void {
+        fn onSpeech(ctx: *anyopaque, event: SpeechEvent) void {
             const self: *@This() = @ptrCast(@alignCast(ctx));
             self.speech_called = true;
-            self.last_text = text;
+            self.last_text = event.text;
+            self.last_amplitude = event.max_amplitude;
         }
 
         fn onError(ctx: *anyopaque, error_code: Error, message: []const u8) void {
@@ -1475,9 +1512,10 @@ test "SpeechEventHandler interface" {
     };
 
     // Test speech callback
-    event_handler.onSpeech("hello world");
+    event_handler.onSpeech(.{ .text = "hello world" });
     try testing.expect(handler.speech_called);
     try testing.expectEqualStrings("hello world", handler.last_text);
+    try testing.expect(handler.last_amplitude == std.math.maxInt(u16));
 
     // Test error callback
     event_handler.onError(Error.AudioDeviceError, "test error");
@@ -1615,10 +1653,10 @@ test "Session session management API" {
         speech_count: u32 = 0,
         error_count: u32 = 0,
 
-        fn onSpeech(ctx: *anyopaque, text: []const u8) void {
+        fn onSpeech(ctx: *anyopaque, event: SpeechEvent) void {
             const self: *@This() = @ptrCast(@alignCast(ctx));
             self.speech_count += 1;
-            _ = text;
+            _ = event;
         }
 
         fn onError(ctx: *anyopaque, error_code: Error, message: []const u8) void {

src/test.zig

@@ -57,9 +57,9 @@ const TestEventHandler = struct {
         self.detailed_error_events.deinit(self.allocator);
     }
 
-    fn onSpeech(ctx: *anyopaque, text: []const u8) void {
+    fn onSpeech(ctx: *anyopaque, event: stt.SpeechEvent) void {
         const self: *TestEventHandler = @ptrCast(@alignCast(ctx));
-        const owned_text = self.allocator.dupe(u8, text) catch return;
+        const owned_text = self.allocator.dupe(u8, event.text) catch return;
         self.speech_events.append(self.allocator, owned_text) catch return;
     }
 
@@ -154,7 +154,7 @@ test "SpeechEventHandler callback invocation" {
     const speech_handler = test_handler.getSpeechEventHandler();
 
     // Test speech callback
-    speech_handler.onSpeech("Hello world");
+    speech_handler.onSpeech(.{ .text = "Hello world" });
     try testing.expect(test_handler.speech_events.items.len == 1);
     try testing.expectEqualStrings("Hello world", test_handler.speech_events.items[0]);
 
@@ -351,13 +351,13 @@ test "Callback error handling robustness" {
     for (0..100) |i| {
         const text = std.fmt.allocPrint(test_allocator, "Speech event {}", .{i}) catch continue;
         defer test_allocator.free(text);
-        speech_handler.onSpeech(text);
+        speech_handler.onSpeech(.{ .text = text });
     }
 
     try testing.expect(test_handler.speech_events.items.len == 100);
 
     // Test mixed callback types
-    speech_handler.onSpeech("Final speech");
+    speech_handler.onSpeech(.{ .text = "Final speech" });
     speech_handler.onError(stt.Error.CallbackError, "Callback error");
 
     const final_error = stt.ErrorInfo.init(stt.Error.InternalError, "Internal error");
@@ -387,7 +387,7 @@ test "Memory management and resource cleanup" {
         defer test_handler.deinit(); // Should not leak memory
 
         const speech_handler = test_handler.getSpeechEventHandler();
-        speech_handler.onSpeech("Test speech");
+        speech_handler.onSpeech(.{ .text = "Test speech" });
         speech_handler.onError(stt.Error.AudioDeviceError, "Test error");
 
         const error_info = stt.ErrorInfo.initWithContext(stt.Error.ModelLoadError, "Test detailed error", "test context");
@@ -461,15 +461,15 @@ test "Complete workflow simulation" {
     try testing.expect(processed_count == audio_samples.len);
 
     // 3. Speech detection phase
-    speech_handler.onSpeech("Hello world");
-    speech_handler.onSpeech("This is a test");
+    speech_handler.onSpeech(.{ .text = "Hello world" });
+    speech_handler.onSpeech(.{ .text = "This is a test" });
 
     // 4. Error handling phase
     const recoverable_error = stt.ErrorInfo.initRecoverable(stt.Error.AudioDeviceBusy, "Audio device temporarily busy", "Retrying in 100ms");
     speech_handler.onDetailedError(recoverable_error);
 
     // 5. Recovery phase
-    speech_handler.onSpeech("Speech recognition resumed");
+    speech_handler.onSpeech(.{ .text = "Speech recognition resumed" });
 
     // 6. Cleanup phase
     const cleanup_info = stt.ErrorInfo.initRecoverable(stt.Error.InternalError, "STT session cleanup completed", "All resources freed");