Refactoring of chrome speech recognition architecture (CL1.3)

content/browser/speech/google_one_shot_remote_engine.cc

Index: content/browser/speech/google_one_shot_remote_engine.cc
diff --git a/content/browser/speech/speech_recognition_request.cc b/content/browser/speech/google_one_shot_remote_engine.cc
similarity index 50%
rename from content/browser/speech/speech_recognition_request.cc
rename to content/browser/speech/google_one_shot_remote_engine.cc
index a14369976d45ff0c267ef0af7221f4cac547fc32..09c2b28d087cb2cdeaa8aebc9d038c9472b1b001 100644
--- a/content/browser/speech/speech_recognition_request.cc
+++ b/content/browser/speech/google_one_shot_remote_engine.cc
@@ -2,16 +2,18 @@
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
-#include "content/browser/speech/speech_recognition_request.h"
+#include "content/browser/speech/google_one_shot_remote_engine.h"
 
 #include <vector>
 
 #include "base/json/json_reader.h"
+#include "base/memory/scoped_ptr.h"
 #include "base/string_number_conversions.h"
 #include "base/string_util.h"
 #include "base/values.h"
 #include "content/browser/speech/audio_buffer.h"
 #include "content/common/net/url_fetcher_impl.h"
+#include "content/public/common/speech_recognition_error.h"
 #include "content/public/common/speech_recognition_result.h"
 #include "net/base/escape.h"
 #include "net/base/load_flags.h"
@@ -19,6 +21,10 @@
 #include "net/url_request/url_request_context_getter.h"
 #include "net/url_request/url_request_status.h"
 
+using content::SpeechRecognitionError;
+using content::SpeechRecognitionHypothesis;
+using content::SpeechRecognitionResult;
+
 namespace {
 
 const char* const kDefaultSpeechRecognitionUrl =
@@ -27,13 +33,20 @@ const char* const kStatusString = "status";
 const char* const kHypothesesString = "hypotheses";
 const char* const kUtteranceString = "utterance";
 const char* const kConfidenceString = "confidence";
-
+const int kWebServiceStatusNoError = 0;
+const int kWebServiceStatusNoSpeech = 4;
+const int kWebServiceStatusNoMatch = 5;
+const int kDefaultConfigSampleRate = 8000;
+const int kDefaultConfigBitsPerSample = 16;
+const speech::AudioEncoder::Codec kDefaultAudioCodec =
+    speech::AudioEncoder::CODEC_FLAC;
 // TODO(satish): Remove this hardcoded value once the page is allowed to
 // set this via an attribute.
 const int kMaxResults = 6;
 
 bool ParseServerResponse(const std::string& response_body,
-                         content::SpeechRecognitionResult* result) {
+                         SpeechRecognitionResult* result,
+                         SpeechRecognitionError* error) {
   if (response_body.empty()) {
     LOG(WARNING) << "ParseServerResponse: Response was empty.";
     return false;
@@ -67,19 +80,21 @@ bool ParseServerResponse(const std::string& response_body,
 
   // Process the status.
   switch (status) {
-  case content::SPEECH_RECOGNITION_ERROR_NONE:
-  case content::SPEECH_RECOGNITION_ERROR_NO_SPEECH:
-  case content::SPEECH_RECOGNITION_ERROR_NO_MATCH:
-    break;
-
-  default:
-    // Other status codes should not be returned by the server.
-    VLOG(1) << "ParseServerResponse: unexpected status code " << status;
-    return false;
+    case kWebServiceStatusNoError:
+      break;
+    case kWebServiceStatusNoSpeech:
+      error->code = content::SPEECH_RECOGNITION_ERROR_NO_SPEECH;
+      return false;
+    case kWebServiceStatusNoMatch:
+      error->code = content::SPEECH_RECOGNITION_ERROR_NO_MATCH;
+      return false;
+    default:
+      error->code = content::SPEECH_RECOGNITION_ERROR_NETWORK;
+      // Other status codes should not be returned by the server.
+      VLOG(1) << "ParseServerResponse: unexpected status code " << status;
+      return false;
   }
 
-  result->error = static_cast<content::SpeechRecognitionErrorCode>(status);
-
   // Get the hypotheses.
   Value* hypotheses_value = NULL;
   if (!response_object->Get(kHypothesesString, &hypotheses_value)) {
@@ -96,6 +111,8 @@ bool ParseServerResponse(const std::string& response_body,
 
   const ListValue* hypotheses_list = static_cast<ListValue*>(hypotheses_value);
 
+  // For now we support only single shot recognition, so we are giving only a
+  // final result, consisting of one fragment (with one or more hypotheses).
   size_t index = 0;
   for (; index < hypotheses_list->GetSize(); ++index) {
     Value* hypothesis = NULL;
@@ -113,6 +130,7 @@ bool ParseServerResponse(const std::string& response_body,
     const DictionaryValue* hypothesis_value =
         static_cast<DictionaryValue*>(hypothesis);
     string16 utterance;
+
     if (!hypothesis_value->GetString(kUtteranceString, &utterance)) {
       LOG(WARNING) << "ParseServerResponse: Missing utterance value.";
       break;
@@ -121,16 +139,14 @@ bool ParseServerResponse(const std::string& response_body,
     // It is not an error if the 'confidence' field is missing.
     double confidence = 0.0;
     hypothesis_value->GetDouble(kConfidenceString, &confidence);
-
-    result->hypotheses.push_back(content::SpeechRecognitionHypothesis(
-        utterance, confidence));
+    result->hypotheses.push_back(SpeechRecognitionHypothesis(utterance,
+                                                             confidence));
   }
 
   if (index < hypotheses_list->GetSize()) {
     result->hypotheses.clear();
     return false;
   }
-
   return true;
 }
 
@@ -138,28 +154,34 @@ bool ParseServerResponse(const std::string& response_body,
 
 namespace speech {
 
-int SpeechRecognitionRequest::url_fetcher_id_for_tests = 0;
+const int GoogleOneShotRemoteEngine::kAudioPacketIntervalMs = 100;
+int GoogleOneShotRemoteEngine::url_fetcher_id_for_tests = 0;
 
-SpeechRecognitionRequest::SpeechRecognitionRequest(
-    net::URLRequestContextGetter* context, Delegate* delegate)
-    : url_context_(context),
-      delegate_(delegate) {
-  DCHECK(delegate);
+GoogleOneShotRemoteEngineConfig::GoogleOneShotRemoteEngineConfig()
+    : filter_profanities(false),
+      audio_sample_rate(kDefaultConfigSampleRate),
+      audio_num_bits_per_sample(kDefaultConfigBitsPerSample) {
 }
 
-SpeechRecognitionRequest::~SpeechRecognitionRequest() {}
+GoogleOneShotRemoteEngineConfig::~GoogleOneShotRemoteEngineConfig() {}
 
-void SpeechRecognitionRequest::Start(const std::string& language,
-                                     const std::string& grammar,
-                                     bool filter_profanities,
-                                     const std::string& hardware_info,
-                                     const std::string& origin_url,
-                                     const std::string& content_type) {
-  DCHECK(!url_fetcher_.get());
+GoogleOneShotRemoteEngine::GoogleOneShotRemoteEngine(
+    net::URLRequestContextGetter* context)
+    : url_context_(context) {
+}
 
-  std::vector<std::string> parts;
+GoogleOneShotRemoteEngine::~GoogleOneShotRemoteEngine() {}
+
+void GoogleOneShotRemoteEngine::SetConfig(
+    const GoogleOneShotRemoteEngineConfig& config) {
+  config_ = config;
+}
+
+void GoogleOneShotRemoteEngine::StartRecognition() {
+  DCHECK(delegate());
+  DCHECK(!url_fetcher_.get());
+  std::string lang_param = config_.language;
 
-  std::string lang_param = language;
   if (lang_param.empty() && url_context_) {
     // If no language is provided then we use the first from the accepted
     // language list. If this list is empty then it defaults to "en-US".
@@ -171,58 +193,108 @@ void SpeechRecognitionRequest::Start(const std::string& language,
     size_t separator = accepted_language_list.find_first_of(",;");
     lang_param = accepted_language_list.substr(0, separator);
   }
+
   if (lang_param.empty())
     lang_param = "en-US";
+
+  std::vector<std::string> parts;
   parts.push_back("lang=" + net::EscapeQueryParamValue(lang_param, true));
 
-  if (!grammar.empty())
-    parts.push_back("lm=" + net::EscapeQueryParamValue(grammar, true));
-  if (!hardware_info.empty())
-    parts.push_back("xhw=" + net::EscapeQueryParamValue(hardware_info, true));
+  if (!config_.grammar.empty())
+    parts.push_back("lm=" + net::EscapeQueryParamValue(config_.grammar, true));
+
+  if (!config_.hardware_info.empty())
+    parts.push_back("xhw=" + net::EscapeQueryParamValue(config_.hardware_info,
+                                                        true));
   parts.push_back("maxresults=" + base::IntToString(kMaxResults));
-  parts.push_back(filter_profanities ? "pfilter=2" : "pfilter=0");
+  parts.push_back(config_.filter_profanities ? "pfilter=2" : "pfilter=0");
 
   GURL url(std::string(kDefaultSpeechRecognitionUrl) + JoinString(parts, '&'));
 
+  encoder_.reset(AudioEncoder::Create(kDefaultAudioCodec,
+                                      config_.audio_sample_rate,
+                                      config_.audio_num_bits_per_sample));
+  DCHECK(encoder_.get());
   url_fetcher_.reset(URLFetcherImpl::Create(url_fetcher_id_for_tests,
                                             url,
                                             URLFetcherImpl::POST,
                                             this));
-  url_fetcher_->SetChunkedUpload(content_type);
+  url_fetcher_->SetChunkedUpload(encoder_->mime_type());
   url_fetcher_->SetRequestContext(url_context_);
-  url_fetcher_->SetReferrer(origin_url);
+  url_fetcher_->SetReferrer(config_.origin_url);
 
   // The speech recognition API does not require user identification as part
   // of requests, so we don't send cookies or auth data for these requests to
   // prevent any accidental connection between users who are logged into the
   // domain for other services (e.g. bookmark sync) with the speech requests.
-  url_fetcher_->SetLoadFlags(
-      net::LOAD_DO_NOT_SAVE_COOKIES | net::LOAD_DO_NOT_SEND_COOKIES |
-      net::LOAD_DO_NOT_SEND_AUTH_DATA);
+  url_fetcher_->SetLoadFlags(net::LOAD_DO_NOT_SAVE_COOKIES |
+                             net::LOAD_DO_NOT_SEND_COOKIES |
+                             net::LOAD_DO_NOT_SEND_AUTH_DATA);
   url_fetcher_->Start();
 }
 
-void SpeechRecognitionRequest::UploadAudioChunk(const AudioChunk& audio_chunk,
-                                                bool is_last_chunk) {
+void GoogleOneShotRemoteEngine::EndRecognition() {
+  url_fetcher_.reset();
+}
+
+void GoogleOneShotRemoteEngine::TakeAudioChunk(const AudioChunk& data) {
+  DCHECK(url_fetcher_.get());
+  DCHECK(encoder_.get());
+  DCHECK_EQ(data.bytes_per_sample(), config_.audio_num_bits_per_sample / 8);
+  encoder_->Encode(data);
+  scoped_ptr<AudioChunk> encoded_data(encoder_->GetEncodedDataAndClear());
+  url_fetcher_->AppendChunkToUpload(encoded_data->AsString(), false);
+}
+
+void GoogleOneShotRemoteEngine::AudioChunksEnded() {
   DCHECK(url_fetcher_.get());
-  url_fetcher_->AppendChunkToUpload(audio_chunk.AsString(), is_last_chunk);
+  DCHECK(encoder_.get());
+
+  // UploadAudioChunk requires a non-empty final buffer. So we encode a packet
+  // of silence in case encoder had no data already.
+  std::vector<int16> samples(
+      config_.audio_sample_rate * kAudioPacketIntervalMs / 1000);
+  AudioChunk dummy_chunk(reinterpret_cast<uint8*>(&samples[0]),
+                         samples.size() * sizeof(int16),
+                         encoder_->bits_per_sample() / 8);
+  encoder_->Encode(dummy_chunk);
+  encoder_->Flush();
+  scoped_ptr<AudioChunk> encoded_dummy_data(encoder_->GetEncodedDataAndClear());
+  DCHECK(!encoded_dummy_data->IsEmpty());
+  encoder_.reset();
+
+  url_fetcher_->AppendChunkToUpload(encoded_dummy_data->AsString(), true);
 }
 
-void SpeechRecognitionRequest::OnURLFetchComplete(
+void GoogleOneShotRemoteEngine::OnURLFetchComplete(
     const content::URLFetcher* source) {
   DCHECK_EQ(url_fetcher_.get(), source);
-
-  content::SpeechRecognitionResult result;
+  SpeechRecognitionResult result;
+  SpeechRecognitionError error(content::SPEECH_RECOGNITION_ERROR_NETWORK);
   std::string data;
-  if (!source->GetStatus().is_success() || source->GetResponseCode() != 200 ||
-      !source->GetResponseAsString(&data) ||
-      !ParseServerResponse(data, &result)) {
-    result.error = content::SPEECH_RECOGNITION_ERROR_NETWORK;
-  }
 
-  DVLOG(1) << "SpeechRecognitionRequest: Invoking delegate with result.";
+  // The default error code in case of parse errors is NETWORK_FAILURE, however
+  // ParseServerResponse can change the error to a more appropriate one.
+  bool error_occurred = (!source->GetStatus().is_success() ||
+                        source->GetResponseCode() != 200 ||
+                        !source->GetResponseAsString(&data) ||
+                        !ParseServerResponse(data, &result, &error));
   url_fetcher_.reset();
-  delegate_->SetRecognitionResult(result);
+  if (error_occurred) {
+    DVLOG(1) << "GoogleOneShotRemoteEngine: Network Error " << error.code;
+    delegate()->OnSpeechRecognitionEngineError(error);
+  } else {
+    DVLOG(1) << "GoogleOneShotRemoteEngine: Invoking delegate with result.";
+    delegate()->OnSpeechRecognitionEngineResult(result);
+  }
+}
+
+bool GoogleOneShotRemoteEngine::IsRecognitionPending() const {
+  return url_fetcher_ != NULL;
+}
+
+int GoogleOneShotRemoteEngine::GetDesiredAudioChunkDurationMs() const {
+  return kAudioPacketIntervalMs;
 }
 
 }  // namespace speech