michael@0: // Copyright (c) 2013 The Chromium Authors. All rights reserved.
michael@0: //
michael@0: // Redistribution and use in source and binary forms, with or without
michael@0: // modification, are permitted provided that the following conditions are
michael@0: // met:
michael@0: //
michael@0: //    * Redistributions of source code must retain the above copyright
michael@0: // notice, this list of conditions and the following disclaimer.
michael@0: //    * Redistributions in binary form must reproduce the above
michael@0: // copyright notice, this list of conditions and the following disclaimer
michael@0: // in the documentation and/or other materials provided with the
michael@0: // distribution.
michael@0: //    * Neither the name of Google Inc. nor the names of its
michael@0: // contributors may be used to endorse or promote products derived from
michael@0: // this software without specific prior written permission.
michael@0: //
michael@0: // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
michael@0: // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
michael@0: // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
michael@0: // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
michael@0: // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
michael@0: // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
michael@0: // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
michael@0: // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
michael@0: // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
michael@0: // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
michael@0: // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
michael@0: 
michael@0: #include "endpointer.h"
michael@0: 
michael@0: #include "AudioSegment.h"
michael@0: 
michael@0: namespace {
michael@0: const int kFrameRate = 200;  // 1 frame = 5ms of audio.
michael@0: }
michael@0: 
michael@0: namespace mozilla {
michael@0: 
michael@0: Endpointer::Endpointer(int sample_rate)
michael@0:     : speech_input_possibly_complete_silence_length_us_(-1),
michael@0:       speech_input_complete_silence_length_us_(-1),
michael@0:       audio_frame_time_us_(0),
michael@0:       sample_rate_(sample_rate),
michael@0:       frame_size_(0) {
michael@0:   Reset();
michael@0: 
michael@0:   frame_size_ = static_cast<int>(sample_rate / static_cast<float>(kFrameRate));
michael@0: 
michael@0:   speech_input_minimum_length_us_ =
michael@0:       static_cast<int64_t>(1.7 * 1000000);
michael@0:   speech_input_complete_silence_length_us_ =
michael@0:       static_cast<int64_t>(0.5 * 1000000);
michael@0:   long_speech_input_complete_silence_length_us_ = -1;
michael@0:   long_speech_length_us_ = -1;
michael@0:   speech_input_possibly_complete_silence_length_us_ =
michael@0:       1 * 1000000;
michael@0: 
michael@0:   // Set the default configuration for Push To Talk mode.
michael@0:   EnergyEndpointerParams ep_config;
michael@0:   ep_config.set_frame_period(1.0f / static_cast<float>(kFrameRate));
michael@0:   ep_config.set_frame_duration(1.0f / static_cast<float>(kFrameRate));
michael@0:   ep_config.set_endpoint_margin(0.2f);
michael@0:   ep_config.set_onset_window(0.15f);
michael@0:   ep_config.set_speech_on_window(0.4f);
michael@0:   ep_config.set_offset_window(0.15f);
michael@0:   ep_config.set_onset_detect_dur(0.09f);
michael@0:   ep_config.set_onset_confirm_dur(0.075f);
michael@0:   ep_config.set_on_maintain_dur(0.10f);
michael@0:   ep_config.set_offset_confirm_dur(0.12f);
michael@0:   ep_config.set_decision_threshold(1000.0f);
michael@0:   ep_config.set_min_decision_threshold(50.0f);
michael@0:   ep_config.set_fast_update_dur(0.2f);
michael@0:   ep_config.set_sample_rate(static_cast<float>(sample_rate));
michael@0:   ep_config.set_min_fundamental_frequency(57.143f);
michael@0:   ep_config.set_max_fundamental_frequency(400.0f);
michael@0:   ep_config.set_contamination_rejection_period(0.25f);
michael@0:   energy_endpointer_.Init(ep_config);
michael@0: }
michael@0: 
michael@0: void Endpointer::Reset() {
michael@0:   old_ep_status_ = EP_PRE_SPEECH;
michael@0:   waiting_for_speech_possibly_complete_timeout_ = false;
michael@0:   waiting_for_speech_complete_timeout_ = false;
michael@0:   speech_previously_detected_ = false;
michael@0:   speech_input_complete_ = false;
michael@0:   audio_frame_time_us_ = 0; // Reset time for packets sent to endpointer.
michael@0:   speech_end_time_us_ = -1;
michael@0:   speech_start_time_us_ = -1;
michael@0: }
michael@0: 
michael@0: void Endpointer::StartSession() {
michael@0:   Reset();
michael@0:   energy_endpointer_.StartSession();
michael@0: }
michael@0: 
michael@0: void Endpointer::EndSession() {
michael@0:   energy_endpointer_.EndSession();
michael@0: }
michael@0: 
michael@0: void Endpointer::SetEnvironmentEstimationMode() {
michael@0:   Reset();
michael@0:   energy_endpointer_.SetEnvironmentEstimationMode();
michael@0: }
michael@0: 
michael@0: void Endpointer::SetUserInputMode() {
michael@0:   energy_endpointer_.SetUserInputMode();
michael@0: }
michael@0: 
michael@0: EpStatus Endpointer::Status(int64_t *time) {
michael@0:   return energy_endpointer_.Status(time);
michael@0: }
michael@0: 
michael@0: EpStatus Endpointer::ProcessAudio(const AudioChunk& raw_audio, float* rms_out) {
michael@0:   MOZ_ASSERT(raw_audio.mBufferFormat == AUDIO_FORMAT_S16, "Audio is not in 16 bit format");
michael@0:   const int16_t* audio_data = static_cast<const int16_t*>(raw_audio.mChannelData[0]);
michael@0:   const int num_samples = raw_audio.mDuration;
michael@0:   EpStatus ep_status = EP_PRE_SPEECH;
michael@0: 
michael@0:   // Process the input data in blocks of frame_size_, dropping any incomplete
michael@0:   // frames at the end (which is ok since typically the caller will be recording
michael@0:   // audio in multiples of our frame size).
michael@0:   int sample_index = 0;
michael@0:   while (sample_index + frame_size_ <= num_samples) {
michael@0:     // Have the endpointer process the frame.
michael@0:     energy_endpointer_.ProcessAudioFrame(audio_frame_time_us_,
michael@0:                                          audio_data + sample_index,
michael@0:                                          frame_size_,
michael@0:                                          rms_out);
michael@0:     sample_index += frame_size_;
michael@0:     audio_frame_time_us_ += (frame_size_ * 1000000) /
michael@0:                          sample_rate_;
michael@0: 
michael@0:     // Get the status of the endpointer.
michael@0:     int64_t ep_time;
michael@0:     ep_status = energy_endpointer_.Status(&ep_time);
michael@0:     if (old_ep_status_ != ep_status)
michael@0:         fprintf(stderr, "Status changed old= %d, new= %d\n", old_ep_status_, ep_status);
michael@0: 
michael@0:     // Handle state changes.
michael@0:     if ((EP_SPEECH_PRESENT == ep_status) &&
michael@0:         (EP_POSSIBLE_ONSET == old_ep_status_)) {
michael@0:       speech_end_time_us_ = -1;
michael@0:       waiting_for_speech_possibly_complete_timeout_ = false;
michael@0:       waiting_for_speech_complete_timeout_ = false;
michael@0:       // Trigger SpeechInputDidStart event on first detection.
michael@0:       if (false == speech_previously_detected_) {
michael@0:         speech_previously_detected_ = true;
michael@0:         speech_start_time_us_ = ep_time;
michael@0:       }
michael@0:     }
michael@0:     if ((EP_PRE_SPEECH == ep_status) &&
michael@0:         (EP_POSSIBLE_OFFSET == old_ep_status_)) {
michael@0:       speech_end_time_us_ = ep_time;
michael@0:       waiting_for_speech_possibly_complete_timeout_ = true;
michael@0:       waiting_for_speech_complete_timeout_ = true;
michael@0:     }
michael@0:     if (ep_time > speech_input_minimum_length_us_) {
michael@0:       // Speech possibly complete timeout.
michael@0:       if ((waiting_for_speech_possibly_complete_timeout_) &&
michael@0:           (ep_time - speech_end_time_us_ >
michael@0:               speech_input_possibly_complete_silence_length_us_)) {
michael@0:         waiting_for_speech_possibly_complete_timeout_ = false;
michael@0:       }
michael@0:       if (waiting_for_speech_complete_timeout_) {
michael@0:         // The length of the silence timeout period can be held constant, or it
michael@0:         // can be changed after a fixed amount of time from the beginning of
michael@0:         // speech.
michael@0:         bool has_stepped_silence =
michael@0:             (long_speech_length_us_ > 0) &&
michael@0:             (long_speech_input_complete_silence_length_us_ > 0);
michael@0:         int64_t requested_silence_length;
michael@0:         if (has_stepped_silence &&
michael@0:             (ep_time - speech_start_time_us_) > long_speech_length_us_) {
michael@0:           requested_silence_length =
michael@0:               long_speech_input_complete_silence_length_us_;
michael@0:         } else {
michael@0:           requested_silence_length =
michael@0:               speech_input_complete_silence_length_us_;
michael@0:         }
michael@0: 
michael@0:         // Speech complete timeout.
michael@0:         if ((ep_time - speech_end_time_us_) > requested_silence_length) {
michael@0:           waiting_for_speech_complete_timeout_ = false;
michael@0:           speech_input_complete_ = true;
michael@0:         }
michael@0:       }
michael@0:     }
michael@0:     old_ep_status_ = ep_status;
michael@0:   }
michael@0:   return ep_status;
michael@0: }
michael@0: 
michael@0: }  // namespace mozilla