lobo/stt
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

report max amplitude in speech output

Browse source
This commit is contained in:
Emil Lerch 2025-10-29 10:44:39 -07:00
parent b1c2ee20ff
commit 4ad75b1b07
Signed by: lobo
GPG key ID: A7B62D657EF764F8
3 changed files with 65 additions and 38 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

20
src/main.zig
View file

@ -26,29 +26,28 @@ const SpeechHandler = struct {
const max_children = 5; const max_children = 5;
const Process = struct { child: ?*std.process.Child, start: i64, id: std.process.Child.Id }; // why id? const Process = struct { child: ?*std.process.Child, start: i64, id: std.process.Child.Id }; // why id?
/// Handle detected speech /// 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 if (builtin.is_test) return; // Suppress output during tests
// Look for noise words and skip it if so // Look for noise words and skip it if so
if (std.mem.eql(u8, text, "huh")) return; if (std.mem.eql(u8, event.text, "huh")) return;
if (std.mem.eql(u8, text, "but")) return; if (std.mem.eql(u8, event.text, "but")) return;
const self: *SpeechHandler = @ptrCast(@alignCast(ctx)); const self: *SpeechHandler = @ptrCast(@alignCast(ctx));
self.speech_count += 1; self.speech_count += 1;
// Print with timestamp for better experience // Print with amplitude for better experience
const timestamp = std.time.timestamp();
var stdout_buffer: [1024]u8 = undefined; var stdout_buffer: [1024]u8 = undefined;
var stdout_writer = std.fs.File.stdout().writer(&stdout_buffer); var stdout_writer = std.fs.File.stdout().writer(&stdout_buffer);
const stdout = &stdout_writer.interface; const stdout = &stdout_writer.interface;
defer stdout.flush() catch std.log.warn("Caught error writing speech data to stdout", .{}); defer stdout.flush() catch std.log.warn("Caught error writing speech data to stdout", .{});
stdout.print("[{}] Speech {}->{?s}: {s}\n", .{ stdout.print("[amp:{}] Speech {}->{?s}: {s}\n", .{
timestamp, event.max_amplitude,
self.speech_count, self.speech_count,
self.exec_program, self.exec_program,
text, event.text,
}) catch std.log.warn("Caught error writing speech data to stdout", .{}); }) catch std.log.warn("Caught error writing speech data to stdout", .{});
// Execute program if specified // Execute program if specified
if (self.exec_program) |program| self.exec(text) catch |err| { if (self.exec_program) |program| self.exec(event.text) catch |err| {
std.log.err("Failed to execute program '{s}': {}", .{ program, err }); std.log.err("Failed to execute program '{s}': {}", .{ program, err });
}; };
} }
@ -526,7 +525,8 @@ test "handler callbacks" {
}; };
// Test that callbacks can be invoked without crashing // Test that callbacks can be invoked without crashing
speech_handler.onSpeech("test speech"); const event = stt.SpeechEvent{ .text = "test speech", .max_amplitude = 500 };
speech_handler.onSpeech(event);
speech_handler.onError(stt.Error.AudioDeviceError, "test error"); speech_handler.onError(stt.Error.AudioDeviceError, "test error");
// If we get here without crashing, the test passes // If we get here without crashing, the test passes

58
src/stt.zig
View file

@ -104,12 +104,20 @@ 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,
};
/// Callback function type for speech detection events /// Callback function type for speech detection events
/// ///
/// Parameters: /// Parameters:
/// - text: Null-terminated string containing the detected speech /// - event: Speech event data containing text and amplitude
/// - user_data: Optional user-provided context data /// - 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 /// Callback function type for error events
/// ///
@ -132,7 +140,7 @@ pub const DetailedErrorCallback = *const fn (error_info: ErrorInfo, user_data: ?
/// with both speech detection and error handling callbacks. /// with both speech detection and error handling callbacks.
pub const SpeechEventHandler = struct { pub const SpeechEventHandler = struct {
/// Function to call when speech is detected /// 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 /// Function to call when an error occurs
onErrorFn: *const fn (ctx: *anyopaque, error_code: Error, message: []const u8) void, onErrorFn: *const fn (ctx: *anyopaque, error_code: Error, message: []const u8) void,
/// Optional function to call for detailed error information /// Optional function to call for detailed error information
@ -141,11 +149,11 @@ pub const SpeechEventHandler = struct {
ctx: *anyopaque, ctx: *anyopaque,
/// Invoke the speech detection callback with error handling /// 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 // Call the speech callback function
// Note: If the callback panics or causes undefined behavior, // Note: If the callback panics or causes undefined behavior,
// there's not much we can do to recover gracefully in Zig // 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 /// Invoke the error callback
@ -974,6 +982,7 @@ pub const Session = struct {
var silence_samples: usize = 0; var silence_samples: usize = 0;
var speech_samples: usize = 0; var speech_samples: usize = 0;
var speech_pos: usize = 0; var speech_pos: usize = 0;
var speech_max_amplitude: u16 = 0;
while (!self.should_stop.load(.acquire)) { while (!self.should_stop.load(.acquire)) {
const available = self.vosk_audio_buffer.available(); const available = self.vosk_audio_buffer.available();
@ -1002,7 +1011,9 @@ pub const Session = struct {
speech_samples = 0; speech_samples = 0;
silence_samples = 0; silence_samples = 0;
speech_pos = 0; speech_pos = 0;
speech_max_amplitude = 0;
} }
if (max_amplitude > speech_max_amplitude) speech_max_amplitude = max_amplitude;
const copy_len = @min(samples_read, speech_buffer.len - speech_pos); 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]); @memcpy(speech_buffer[speech_pos .. speech_pos + copy_len], vosk_buffer[0..copy_len]);
speech_pos += copy_len; speech_pos += copy_len;
@ -1016,7 +1027,7 @@ pub const Session = struct {
if (silence_samples >= silence_duration_ms * samples_per_ms) { if (silence_samples >= silence_duration_ms * samples_per_ms) {
if (speech_samples >= min_speech_duration_ms * samples_per_ms) { if (speech_samples >= min_speech_duration_ms * samples_per_ms) {
self.processVoskAudio(speech_buffer[0..speech_pos]) catch |err| { self.processVoskAudio(speech_buffer[0..speech_pos], speech_max_amplitude) catch |err| {
std.log.err("Error processing speech: {}", .{err}); std.log.err("Error processing speech: {}", .{err});
}; };
} }
@ -1025,6 +1036,7 @@ pub const Session = struct {
speech_pos = 0; speech_pos = 0;
speech_samples = 0; speech_samples = 0;
silence_samples = 0; silence_samples = 0;
speech_max_amplitude = 0;
if (self.vosk_recognizer) |rec| { if (self.vosk_recognizer) |rec| {
c.vosk_recognizer_reset(rec); c.vosk_recognizer_reset(rec);
@ -1038,7 +1050,7 @@ pub const Session = struct {
} }
/// Process complete audio clip with Vosk and handle results /// Process complete audio clip with Vosk and handle results
fn processVoskAudio(self: *Session, audio_data: []const i16) !void { fn processVoskAudio(self: *Session, audio_data: []const i16, max_amplitude: u16) !void {
if (self.vosk_recognizer == null) return Error.InvalidState; if (self.vosk_recognizer == null) return Error.InvalidState;
const audio_bytes = std.mem.sliceAsBytes(audio_data); const audio_bytes = std.mem.sliceAsBytes(audio_data);
@ -1048,14 +1060,14 @@ pub const Session = struct {
const result_cstr = c.vosk_recognizer_final_result(self.vosk_recognizer); const result_cstr = c.vosk_recognizer_final_result(self.vosk_recognizer);
if (result_cstr != null) { if (result_cstr != null) {
const result_str = std.mem.span(result_cstr); const result_str = std.mem.span(result_cstr);
self.parseVoskResult(result_str) catch |err| { self.parseVoskResult(result_str, max_amplitude) catch |err| {
self.options.event_handler.onError(err, "Failed to parse Vosk result"); self.options.event_handler.onError(err, "Failed to parse Vosk result");
}; };
} }
} }
/// Parse Vosk JSON result and extract recognized text /// 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, max_amplitude: u16) !void {
if (json_str.len == 0) return; if (json_str.len == 0) return;
const text_key = "\"text\""; const text_key = "\"text\"";
@ -1074,7 +1086,11 @@ pub const Session = struct {
if (std.mem.indexOf(u8, json_str[s..], "\"")) |quote_end| { if (std.mem.indexOf(u8, json_str[s..], "\"")) |quote_end| {
const text = json_str[s .. s + quote_end]; const text = json_str[s .. s + quote_end];
if (text.len > 0 and !std.mem.eql(u8, text, " ")) { if (text.len > 0 and !std.mem.eql(u8, text, " ")) {
self.options.event_handler.onSpeech(text); const event = SpeechEvent{
.text = text,
.max_amplitude = max_amplitude,
};
self.options.event_handler.onSpeech(event);
} }
} }
} }
@ -1376,9 +1392,9 @@ test "Options validation" {
// Test valid options // Test valid options
const DummyHandler = struct { const DummyHandler = struct {
fn onSpeech(ctx: *anyopaque, text: []const u8) void { fn onSpeech(ctx: *anyopaque, event: SpeechEvent) void {
_ = ctx; _ = ctx;
_ = text; _ = event;
} }
fn onError(ctx: *anyopaque, error_code: Error, message: []const u8) void { fn onError(ctx: *anyopaque, error_code: Error, message: []const u8) void {
_ = ctx; _ = ctx;
@ -1412,9 +1428,9 @@ test "Session state management" {
const testing = std.testing; const testing = std.testing;
const DummyHandler = struct { const DummyHandler = struct {
fn onSpeech(ctx: *anyopaque, text: []const u8) void { fn onSpeech(ctx: *anyopaque, event: SpeechEvent) void {
_ = ctx; _ = ctx;
_ = text; _ = event;
} }
fn onError(ctx: *anyopaque, error_code: Error, message: []const u8) void { fn onError(ctx: *anyopaque, error_code: Error, message: []const u8) void {
_ = ctx; _ = ctx;
@ -1451,12 +1467,14 @@ test "SpeechEventHandler interface" {
speech_called: bool = false, speech_called: bool = false,
error_called: bool = false, error_called: bool = false,
last_text: []const u8 = "", last_text: []const u8 = "",
last_amplitude: u16 = 0,
last_error: Error = Error.InitializationFailed, 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)); const self: *@This() = @ptrCast(@alignCast(ctx));
self.speech_called = true; 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 { fn onError(ctx: *anyopaque, error_code: Error, message: []const u8) void {
@ -1475,9 +1493,11 @@ test "SpeechEventHandler interface" {
}; };
// Test speech callback // Test speech callback
event_handler.onSpeech("hello world"); const event = SpeechEvent{ .text = "hello world", .max_amplitude = 1234 };
event_handler.onSpeech(event);
try testing.expect(handler.speech_called); try testing.expect(handler.speech_called);
try testing.expectEqualStrings("hello world", handler.last_text); try testing.expectEqualStrings("hello world", handler.last_text);
try testing.expect(handler.last_amplitude == 1234);
// Test error callback // Test error callback
event_handler.onError(Error.AudioDeviceError, "test error"); event_handler.onError(Error.AudioDeviceError, "test error");
@ -1615,10 +1635,10 @@ test "Session session management API" {
speech_count: u32 = 0, speech_count: u32 = 0,
error_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)); const self: *@This() = @ptrCast(@alignCast(ctx));
self.speech_count += 1; self.speech_count += 1;
_ = text; _ = event;
} }
fn onError(ctx: *anyopaque, error_code: Error, message: []const u8) void { fn onError(ctx: *anyopaque, error_code: Error, message: []const u8) void {

25
src/test.zig
View file

@ -57,9 +57,9 @@ const TestEventHandler = struct {
self.detailed_error_events.deinit(self.allocator); 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 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; self.speech_events.append(self.allocator, owned_text) catch return;
} }
@ -154,7 +154,8 @@ test "SpeechEventHandler callback invocation" {
const speech_handler = test_handler.getSpeechEventHandler(); const speech_handler = test_handler.getSpeechEventHandler();
// Test speech callback // Test speech callback
speech_handler.onSpeech("Hello world"); const event = stt.SpeechEvent{ .text = "Hello world", .max_amplitude = 1000 };
speech_handler.onSpeech(event);
try testing.expect(test_handler.speech_events.items.len == 1); try testing.expect(test_handler.speech_events.items.len == 1);
try testing.expectEqualStrings("Hello world", test_handler.speech_events.items[0]); try testing.expectEqualStrings("Hello world", test_handler.speech_events.items[0]);
@ -351,13 +352,15 @@ test "Callback error handling robustness" {
for (0..100) |i| { for (0..100) |i| {
const text = std.fmt.allocPrint(test_allocator, "Speech event {}", .{i}) catch continue; const text = std.fmt.allocPrint(test_allocator, "Speech event {}", .{i}) catch continue;
defer test_allocator.free(text); defer test_allocator.free(text);
speech_handler.onSpeech(text); const event = stt.SpeechEvent{ .text = text, .max_amplitude = @intCast(i) };
speech_handler.onSpeech(event);
} }
try testing.expect(test_handler.speech_events.items.len == 100); try testing.expect(test_handler.speech_events.items.len == 100);
// Test mixed callback types // Test mixed callback types
speech_handler.onSpeech("Final speech"); const final_event = stt.SpeechEvent{ .text = "Final speech", .max_amplitude = 800 };
speech_handler.onSpeech(final_event);
speech_handler.onError(stt.Error.CallbackError, "Callback error"); speech_handler.onError(stt.Error.CallbackError, "Callback error");
const final_error = stt.ErrorInfo.init(stt.Error.InternalError, "Internal error"); const final_error = stt.ErrorInfo.init(stt.Error.InternalError, "Internal error");
@ -387,7 +390,8 @@ test "Memory management and resource cleanup" {
defer test_handler.deinit(); // Should not leak memory defer test_handler.deinit(); // Should not leak memory
const speech_handler = test_handler.getSpeechEventHandler(); const speech_handler = test_handler.getSpeechEventHandler();
speech_handler.onSpeech("Test speech"); const event = stt.SpeechEvent{ .text = "Test speech", .max_amplitude = 500 };
speech_handler.onSpeech(event);
speech_handler.onError(stt.Error.AudioDeviceError, "Test error"); speech_handler.onError(stt.Error.AudioDeviceError, "Test error");
const error_info = stt.ErrorInfo.initWithContext(stt.Error.ModelLoadError, "Test detailed error", "test context"); const error_info = stt.ErrorInfo.initWithContext(stt.Error.ModelLoadError, "Test detailed error", "test context");
@ -461,15 +465,18 @@ test "Complete workflow simulation" {
try testing.expect(processed_count == audio_samples.len); try testing.expect(processed_count == audio_samples.len);
// 3. Speech detection phase // 3. Speech detection phase
speech_handler.onSpeech("Hello world"); const event1 = stt.SpeechEvent{ .text = "Hello world", .max_amplitude = 2000 };
speech_handler.onSpeech("This is a test"); speech_handler.onSpeech(event1);
const event2 = stt.SpeechEvent{ .text = "This is a test", .max_amplitude = 1500 };
speech_handler.onSpeech(event2);
// 4. Error handling phase // 4. Error handling phase
const recoverable_error = stt.ErrorInfo.initRecoverable(stt.Error.AudioDeviceBusy, "Audio device temporarily busy", "Retrying in 100ms"); const recoverable_error = stt.ErrorInfo.initRecoverable(stt.Error.AudioDeviceBusy, "Audio device temporarily busy", "Retrying in 100ms");
speech_handler.onDetailedError(recoverable_error); speech_handler.onDetailedError(recoverable_error);
// 5. Recovery phase // 5. Recovery phase
speech_handler.onSpeech("Speech recognition resumed"); const resume_event = stt.SpeechEvent{ .text = "Speech recognition resumed", .max_amplitude = 1800 };
speech_handler.onSpeech(resume_event);
// 6. Cleanup phase // 6. Cleanup phase
const cleanup_info = stt.ErrorInfo.initRecoverable(stt.Error.InternalError, "STT session cleanup completed", "All resources freed"); const cleanup_info = stt.ErrorInfo.initRecoverable(stt.Error.InternalError, "STT session cleanup completed", "All resources freed");