Split iSAC encoder/decoder: Test more cases (and make sure they work)

webrtc/modules/audio_coding/codecs/isac/unittest.cc

Index: webrtc/modules/audio_coding/codecs/isac/unittest.cc
diff --git a/webrtc/modules/audio_coding/codecs/isac/unittest.cc b/webrtc/modules/audio_coding/codecs/isac/unittest.cc
index a80fd08bcfcb33b8d8eaaa88d6d609386802ed06..d05ffa6e48f646767d709052ef4a2ea9dda3eee5 100644
--- a/webrtc/modules/audio_coding/codecs/isac/unittest.cc
+++ b/webrtc/modules/audio_coding/codecs/isac/unittest.cc
@@ -24,10 +24,11 @@ namespace webrtc {
 
 namespace {
 
+const int kIsacNumberOfSamples = 32 * 60;  // 60 ms at 32 kHz
+
 std::vector<int16_t> LoadSpeechData() {
   webrtc::test::InputAudioFile input_file(
       webrtc::test::ResourcePath("audio_coding/testfile32kHz", "pcm"));
-  static const int kIsacNumberOfSamples = 32 * 60;  // 60 ms at 32 kHz
   std::vector<int16_t> speech_data(kIsacNumberOfSamples);
   input_file.Read(kIsacNumberOfSamples, speech_data.data());
   return speech_data;
@@ -41,32 +42,45 @@ IsacBandwidthInfo GetBwInfo(typename T::instance_type* inst) {
   return bi;
 }
 
+// Encodes one packet. Returns the packet duration in milliseconds.
 template <typename T>
-rtc::Buffer EncodePacket(typename T::instance_type* inst,
-                         const IsacBandwidthInfo* bi,
-                         const int16_t* speech_data,
-                         int framesize_ms) {
-  rtc::Buffer output(1000);
-  for (int i = 0;; ++i) {
+int EncodePacket(typename T::instance_type* inst,
+                 const IsacBandwidthInfo* bi,
+                 const int16_t* speech_data,
+                 rtc::Buffer* output) {
+  output->SetSize(1000);
+  for (int duration_ms = 10;; duration_ms += 10) {
     if (bi)
       T::SetBandwidthInfo(inst, bi);
-    int encoded_bytes = T::Encode(inst, speech_data, output.data());
-    if (i + 1 == framesize_ms / 10) {
+    int encoded_bytes = T::Encode(inst, speech_data, output->data());
+    if (encoded_bytes > 0 || duration_ms >= 60) {
       EXPECT_GT(encoded_bytes, 0);
-      EXPECT_LE(static_cast<size_t>(encoded_bytes), output.size());
-      output.SetSize(encoded_bytes);
-      return output;
+      EXPECT_LE(static_cast<size_t>(encoded_bytes), output->size());
+      output->SetSize(encoded_bytes);
+      return duration_ms;
     }
-    EXPECT_EQ(0, encoded_bytes);
   }
 }
 
+template <typename T>
+std::vector<int16_t> DecodePacket(typename T::instance_type* inst,
+                                  const rtc::Buffer& encoded) {
+  std::vector<int16_t> decoded(kIsacNumberOfSamples);
+  int16_t speech_type;
+  int nsamples = T::DecodeInternal(inst, encoded.data(), encoded.size(),
+                                   &decoded.front(), &speech_type);
+  EXPECT_GT(nsamples, 0);
+  EXPECT_LE(static_cast<size_t>(nsamples), decoded.size());
+  decoded.resize(nsamples);
+  return decoded;
+}
+
 class BoundedCapacityChannel final {
  public:
-  BoundedCapacityChannel(int rate_bits_per_second)
+  BoundedCapacityChannel(int sample_rate_hz, int rate_bits_per_second)
       : current_time_rtp_(0),
         channel_rate_bytes_per_sample_(rate_bits_per_second /
-                                       (8.0 * kSamplesPerSecond)) {}
+                                       (8.0 * sample_rate_hz)) {}
 
   // Simulate sending the given number of bytes at the given RTP time. Returns
   // the new current RTP time after the sending is done.
@@ -81,47 +95,6 @@ class BoundedCapacityChannel final {
   // The somewhat strange unit for channel rate, bytes per sample, is because
   // RTP time is measured in samples:
   const double channel_rate_bytes_per_sample_;
-  static const int kSamplesPerSecond = 16000;
-};
-
-template <typename T, bool adaptive>
-struct TestParam {};
-
-template <>
-struct TestParam<IsacFloat, true> {
-  static const int time_to_settle = 200;
-  static int ExpectedRateBitsPerSecond(int rate_bits_per_second) {
-    return rate_bits_per_second;
-  }
-};
-
-template <>
-struct TestParam<IsacFix, true> {
-  static const int time_to_settle = 350;
-  static int ExpectedRateBitsPerSecond(int rate_bits_per_second) {
-    // For some reason, IsacFix fails to adapt to the channel's actual
-    // bandwidth. Instead, it settles on a few hundred packets at 10kbit/s,
-    // then a few hundred at 5kbit/s, then a few hundred at 10kbit/s, and so
-    // on. The 200 packets starting at 350 are in the middle of the first
-    // 10kbit/s run.
-    return 10000;
-  }
-};
-
-template <>
-struct TestParam<IsacFloat, false> {
-  static const int time_to_settle = 0;
-  static int ExpectedRateBitsPerSecond(int rate_bits_per_second) {
-    return 32000;
-  }
-};
-
-template <>
-struct TestParam<IsacFix, false> {
-  static const int time_to_settle = 0;
-  static int ExpectedRateBitsPerSecond(int rate_bits_per_second) {
-    return 16000;
-  }
 };
 
 // Test that the iSAC encoder produces identical output whether or not we use a
@@ -129,143 +102,153 @@ struct TestParam<IsacFix, false> {
 // communicate BW estimation info explicitly.
 template <typename T, bool adaptive>
 void TestGetSetBandwidthInfo(const int16_t* speech_data,
-                             int rate_bits_per_second) {
-  using Param = TestParam<T, adaptive>;
-  const int framesize_ms = adaptive ? 60 : 30;
+                             int rate_bits_per_second,
+                             int sample_rate_hz,
+                             int frame_size_ms) {
+  const int bit_rate = 32000;
 
   // Conjoined encoder/decoder pair:
   typename T::instance_type* encdec;
   ASSERT_EQ(0, T::Create(&encdec));
   ASSERT_EQ(0, T::EncoderInit(encdec, adaptive ? 0 : 1));
   ASSERT_EQ(0, T::DecoderInit(encdec));
+  ASSERT_EQ(0, T::SetEncSampRate(encdec, sample_rate_hz));
+  if (adaptive)
+    ASSERT_EQ(0, T::ControlBwe(encdec, bit_rate, frame_size_ms, false));
+  else
+    ASSERT_EQ(0, T::Control(encdec, bit_rate, frame_size_ms));
 
   // Disjoint encoder/decoder pair:
   typename T::instance_type* enc;
   ASSERT_EQ(0, T::Create(&enc));
   ASSERT_EQ(0, T::EncoderInit(enc, adaptive ? 0 : 1));
+  ASSERT_EQ(0, T::SetEncSampRate(enc, sample_rate_hz));
+  if (adaptive)
+    ASSERT_EQ(0, T::ControlBwe(enc, bit_rate, frame_size_ms, false));
+  else
+    ASSERT_EQ(0, T::Control(enc, bit_rate, frame_size_ms));
   typename T::instance_type* dec;
   ASSERT_EQ(0, T::Create(&dec));
   ASSERT_EQ(0, T::DecoderInit(dec));
+  T::SetInitialBweBottleneck(dec, bit_rate);
+  T::SetEncSampRateInDecoder(dec, sample_rate_hz);
 
   // 0. Get initial BW info from decoder.
   auto bi = GetBwInfo<T>(dec);
 
-  BoundedCapacityChannel channel1(rate_bits_per_second),
-      channel2(rate_bits_per_second);
-  std::vector<size_t> packet_sizes;
-  for (int i = 0; i < Param::time_to_settle + 200; ++i) {
+  BoundedCapacityChannel channel1(sample_rate_hz, rate_bits_per_second),
+      channel2(sample_rate_hz, rate_bits_per_second);
+
+  int elapsed_time_ms = 0;
+  for (int i = 0; elapsed_time_ms < 10000; ++i) {
     std::ostringstream ss;
     ss << " i = " << i;
     SCOPED_TRACE(ss.str());
 
-    // 1. Encode 6 * 10 ms (adaptive) or 3 * 10 ms (nonadaptive). The separate
-    // encoder is given the BW info before each encode call.
-    auto bitstream1 =
-        EncodePacket<T>(encdec, nullptr, speech_data, framesize_ms);
-    auto bitstream2 = EncodePacket<T>(enc, &bi, speech_data, framesize_ms);
+    // 1. Encode 3 * 10 ms or 6 * 10 ms. The separate encoder is given the BW
+    // info before each encode call.
+    rtc::Buffer bitstream1, bitstream2;
+    int duration1_ms =
+        EncodePacket<T>(encdec, nullptr, speech_data, &bitstream1);
+    int duration2_ms = EncodePacket<T>(enc, &bi, speech_data, &bitstream2);
+    EXPECT_EQ(duration1_ms, duration2_ms);
+    if (adaptive)
+      EXPECT_TRUE(duration1_ms == 30 || duration1_ms == 60);
+    else
+      EXPECT_EQ(frame_size_ms, duration1_ms);
+    ASSERT_EQ(bitstream1.size(), bitstream2.size());
     EXPECT_EQ(bitstream1, bitstream2);
-    if (i > Param::time_to_settle)
-      packet_sizes.push_back(bitstream1.size());
-
-    // 2. Deliver the encoded data to the decoders (but don't actually ask them
-    // to decode it; that's not necessary). Then get new BW info from the
-    // separate decoder.
-    const int samples_per_packet = 16 * framesize_ms;
-    const int send_time = i * samples_per_packet;
+
+    // 2. Deliver the encoded data to the decoders.
+    const int send_time = elapsed_time_ms * (sample_rate_hz / 1000);
     EXPECT_EQ(0, T::UpdateBwEstimate(
                      encdec, bitstream1.data(), bitstream1.size(), i, send_time,
                      channel1.Send(send_time, bitstream1.size())));
     EXPECT_EQ(0, T::UpdateBwEstimate(
                      dec, bitstream2.data(), bitstream2.size(), i, send_time,
                      channel2.Send(send_time, bitstream2.size())));
+
+    // 3. Decode, and get new BW info from the separate decoder.
+    ASSERT_EQ(0, T::SetDecSampRate(encdec, sample_rate_hz));
+    ASSERT_EQ(0, T::SetDecSampRate(dec, sample_rate_hz));
+    auto decoded1 = DecodePacket<T>(encdec, bitstream1);
+    auto decoded2 = DecodePacket<T>(dec, bitstream2);
+    EXPECT_EQ(decoded1, decoded2);
     bi = GetBwInfo<T>(dec);
+
+    elapsed_time_ms += duration1_ms;
   }
 
   EXPECT_EQ(0, T::Free(encdec));
   EXPECT_EQ(0, T::Free(enc));
   EXPECT_EQ(0, T::Free(dec));
-
-  // The average send bitrate is close to the channel's capacity.
-  double avg_size =
-      std::accumulate(packet_sizes.begin(), packet_sizes.end(), 0) /
-      static_cast<double>(packet_sizes.size());
-  double avg_rate_bits_per_second = 8.0 * avg_size / (framesize_ms * 1e-3);
-  double expected_rate_bits_per_second =
-      Param::ExpectedRateBitsPerSecond(rate_bits_per_second);
-  EXPECT_GT(avg_rate_bits_per_second / expected_rate_bits_per_second, 0.95);
-  EXPECT_LT(avg_rate_bits_per_second / expected_rate_bits_per_second, 1.06);
-
-  // The largest packet isn't that large, and the smallest not that small.
-  size_t min_size = *std::min_element(packet_sizes.begin(), packet_sizes.end());
-  size_t max_size = *std::max_element(packet_sizes.begin(), packet_sizes.end());
-  double size_range = max_size - min_size;
-  EXPECT_LE(size_range / avg_size, 0.16);
 }
 
-}  // namespace
-
-TEST(IsacCommonTest, GetSetBandwidthInfoFloat12kAdaptive) {
-  TestGetSetBandwidthInfo<IsacFloat, true>(LoadSpeechData().data(), 12000);
-}
+enum class IsacType { Fix, Float };
 
-TEST(IsacCommonTest, GetSetBandwidthInfoFloat15kAdaptive) {
-  TestGetSetBandwidthInfo<IsacFloat, true>(LoadSpeechData().data(), 15000);
+std::ostream& operator<<(std::ostream& os, IsacType t) {
+  os << (t == IsacType::Fix ? "fix" : "float");
+  return os;
 }
 
-TEST(IsacCommonTest, GetSetBandwidthInfoFloat19kAdaptive) {
-  TestGetSetBandwidthInfo<IsacFloat, true>(LoadSpeechData().data(), 19000);
-}
+struct IsacTestParam {
+  IsacType isac_type;
+  bool adaptive;
+  int channel_rate_bits_per_second;
+  int sample_rate_hz;
+  int frame_size_ms;
 
-TEST(IsacCommonTest, GetSetBandwidthInfoFloat22kAdaptive) {
-  TestGetSetBandwidthInfo<IsacFloat, true>(LoadSpeechData().data(), 22000);
-}
-
-TEST(IsacCommonTest, GetSetBandwidthInfoFix12kAdaptive) {
-  TestGetSetBandwidthInfo<IsacFix, true>(LoadSpeechData().data(), 12000);
-}
-
-TEST(IsacCommonTest, GetSetBandwidthInfoFix15kAdaptive) {
-  TestGetSetBandwidthInfo<IsacFix, true>(LoadSpeechData().data(), 15000);
-}
-
-TEST(IsacCommonTest, GetSetBandwidthInfoFix19kAdaptive) {
-  TestGetSetBandwidthInfo<IsacFix, true>(LoadSpeechData().data(), 19000);
-}
-
-TEST(IsacCommonTest, GetSetBandwidthInfoFix22kAdaptive) {
-  TestGetSetBandwidthInfo<IsacFix, true>(LoadSpeechData().data(), 22000);
-}
-
-TEST(IsacCommonTest, GetSetBandwidthInfoFloat12k) {
-  TestGetSetBandwidthInfo<IsacFloat, false>(LoadSpeechData().data(), 12000);
-}
-
-TEST(IsacCommonTest, GetSetBandwidthInfoFloat15k) {
-  TestGetSetBandwidthInfo<IsacFloat, false>(LoadSpeechData().data(), 15000);
-}
-
-TEST(IsacCommonTest, GetSetBandwidthInfoFloat19k) {
-  TestGetSetBandwidthInfo<IsacFloat, false>(LoadSpeechData().data(), 19000);
-}
-
-TEST(IsacCommonTest, GetSetBandwidthInfoFloat22k) {
-  TestGetSetBandwidthInfo<IsacFloat, false>(LoadSpeechData().data(), 22000);
-}
-
-TEST(IsacCommonTest, GetSetBandwidthInfoFix12k) {
-  TestGetSetBandwidthInfo<IsacFix, false>(LoadSpeechData().data(), 12000);
-}
+  friend std::ostream& operator<<(std::ostream& os, const IsacTestParam& itp) {
+    os << '{' << itp.isac_type << ','
+       << (itp.adaptive ? "adaptive" : "nonadaptive") << ','
+       << itp.channel_rate_bits_per_second << ',' << itp.sample_rate_hz << ','
+       << itp.frame_size_ms << '}';
+    return os;
+  }
+};
 
-TEST(IsacCommonTest, GetSetBandwidthInfoFix15k) {
-  TestGetSetBandwidthInfo<IsacFix, false>(LoadSpeechData().data(), 15000);
-}
+class IsacCommonTest : public testing::TestWithParam<IsacTestParam> {};
 
-TEST(IsacCommonTest, GetSetBandwidthInfoFix19k) {
-  TestGetSetBandwidthInfo<IsacFix, false>(LoadSpeechData().data(), 19000);
-}
+}  // namespace
 
-TEST(IsacCommonTest, GetSetBandwidthInfoFix22k) {
-  TestGetSetBandwidthInfo<IsacFix, false>(LoadSpeechData().data(), 22000);
-}
+TEST_P(IsacCommonTest, GetSetBandwidthInfo) {
+  auto p = GetParam();
+  auto test_fun = [p] {
+    if (p.isac_type == IsacType::Fix) {
+      if (p.adaptive)
+        return TestGetSetBandwidthInfo<IsacFix, true>;
+      else
+        return TestGetSetBandwidthInfo<IsacFix, false>;
+    } else {
+      if (p.adaptive)
+        return TestGetSetBandwidthInfo<IsacFloat, true>;
+      else
+        return TestGetSetBandwidthInfo<IsacFloat, false>;
+    }
+  }();
+  test_fun(LoadSpeechData().data(), p.channel_rate_bits_per_second,
+           p.sample_rate_hz, p.frame_size_ms);
+}
+
+std::vector<IsacTestParam> TestCases() {
+  static const IsacType types[] = {IsacType::Fix, IsacType::Float};
+  static const bool adaptives[] = {true, false};
+  static const int channel_rates[] = {12000, 15000, 19000, 22000};
+  static const int sample_rates[] = {16000, 32000};
+  static const int frame_sizes[] = {30, 60};
+  std::vector<IsacTestParam> cases;
+  for (IsacType type : types)
+    for (bool adaptive : adaptives)
+      for (int channel_rate : channel_rates)
+        for (int sample_rate : sample_rates)
+          if (!(type == IsacType::Fix && sample_rate == 32000))
+            for (int frame_size : frame_sizes)
+              if (!(sample_rate == 32000 && frame_size == 60))
+                cases.push_back(
+                    {type, adaptive, channel_rate, sample_rate, frame_size});
+  return cases;
+}
+
+INSTANTIATE_TEST_CASE_P(, IsacCommonTest, testing::ValuesIn(TestCases()));
 
 }  // namespace webrtc