Add G722 decoder with pure python implementation

bumble/decoder.py

@@ -0,0 +1,428 @@
+# Copyright 2023 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# -----------------------------------------------------------------------------
+# Constants
+# -----------------------------------------------------------------------------
+# fmt: off
+
+WL = [-60, -30, 58, 172, 334, 538, 1198, 3042]
+RL42 = [0, 7, 6, 5, 4, 3, 2, 1, 7, 6, 5, 4, 3, 2, 1, 0]
+ILB = [
+    2048,
+    2093,
+    2139,
+    2186,
+    2233,
+    2282,
+    2332,
+    2383,
+    2435,
+    2489,
+    2543,
+    2599,
+    2656,
+    2714,
+    2774,
+    2834,
+    2896,
+    2960,
+    3025,
+    3091,
+    3158,
+    3228,
+    3298,
+    3371,
+    3444,
+    3520,
+    3597,
+    3676,
+    3756,
+    3838,
+    3922,
+    4008,
+]
+WH = [0, -214, 798]
+RH2 = [2, 1, 2, 1]
+# Values in QM2/QM4/QM6 left shift three bits than original g722 specification.
+QM2 = [-7408, -1616, 7408, 1616]
+QM4 = [
+    0,
+    -20456,
+    -12896,
+    -8968,
+    -6288,
+    -4240,
+    -2584,
+    -1200,
+    20456,
+    12896,
+    8968,
+    6288,
+    4240,
+    2584,
+    1200,
+    0,
+]
+QM6 = [
+    -136,
+    -136,
+    -136,
+    -136,
+    -24808,
+    -21904,
+    -19008,
+    -16704,
+    -14984,
+    -13512,
+    -12280,
+    -11192,
+    -10232,
+    -9360,
+    -8576,
+    -7856,
+    -7192,
+    -6576,
+    -6000,
+    -5456,
+    -4944,
+    -4464,
+    -4008,
+    -3576,
+    -3168,
+    -2776,
+    -2400,
+    -2032,
+    -1688,
+    -1360,
+    -1040,
+    -728,
+    24808,
+    21904,
+    19008,
+    16704,
+    14984,
+    13512,
+    12280,
+    11192,
+    10232,
+    9360,
+    8576,
+    7856,
+    7192,
+    6576,
+    6000,
+    5456,
+    4944,
+    4464,
+    4008,
+    3576,
+    3168,
+    2776,
+    2400,
+    2032,
+    1688,
+    1360,
+    1040,
+    728,
+    432,
+    136,
+    -432,
+    -136,
+]
+QMF_COEFFS = [3, -11, 12, 32, -210, 951, 3876, -805, 362, -156, 53, -11]
+
+# fmt: on
+
+
+# -----------------------------------------------------------------------------
+# Classes
+# -----------------------------------------------------------------------------
+class G722Decoder(object):
+    """G.722 decoder with bitrate 64kbit/s.
+
+    For the Blocks in the sub-band decoders, please refer to the G.722
+    specification for the required information. G722 specification:
+    https://www.itu.int/rec/T-REC-G.722-201209-I
+    """
+
+    def __init__(self):
+        self._x = [0] * 24
+        self._band = [Band(), Band()]
+        # The initial value in BLOCK 3L
+        self._band[0].det = 32
+        # The initial value in BLOCK 3H
+        self._band[1].det = 8
+
+    def decode_frame(self, encoded_data) -> bytearray:
+        result_array = bytearray(len(encoded_data) * 4)
+        self.g722_decode(result_array, encoded_data)
+        return result_array
+
+    def g722_decode(self, result_array, encoded_data) -> int:
+        """Decode the data frame using g722 decoder."""
+        rlow = 0
+        rhigh = 0
+        xout1 = 0
+        xout2 = 0
+        code = 0
+        result_length = 0
+        j = 0
+
+        for j in range(0, len(encoded_data)):
+            code = encoded_data[j]
+            higher_bits = (code >> 6) & 0x03
+            lower_bits = code & 0x3F
+
+            rlow = self.lower_sub_band_decoder(lower_bits)
+            rhigh = self.higher_sub_band_decoder(higher_bits)
+
+            # Apply the receive QMF
+            for i in range(22):
+                self._x[i] = self._x[i + 2]
+
+            self._x[22] = rlow + rhigh
+            self._x[23] = rlow - rhigh
+            xout1 = 0
+            xout2 = 0
+
+            for i in range(12):
+                xout2 += self._x[2 * i] * QMF_COEFFS[i]
+                xout1 += self._x[2 * i + 1] * QMF_COEFFS[11 - i]
+
+            result_length = self.update_decoded_result(
+                xout1, result_length, result_array
+            )
+            result_length = self.update_decoded_result(
+                xout2, result_length, result_array
+            )
+
+        return result_length
+
+    def update_decoded_result(self, xout, byte_length, byte_array) -> int:
+        result = (int)(xout >> 11)
+        bytes_result = result.to_bytes(2, 'little', signed=True)
+        byte_array[byte_length] = bytes_result[0]
+        byte_array[byte_length + 1] = bytes_result[1]
+        return byte_length + 2
+
+    def lower_sub_band_decoder(self, lower_bits) -> int:
+        """Lower sub-band decoder for last six bits."""
+
+        # Block 5L
+        # INVQBL
+        wd1 = lower_bits
+        wd2 = QM6[wd1]
+        wd1 >>= 2
+        wd2 = (self._band[0].det * wd2) >> 15
+        # RECONS
+        rlow = self._band[0].s + wd2
+
+        # Block 6L
+        # LIMIT
+        if rlow > 16383:
+            rlow = 16383
+        elif rlow < -16384:
+            rlow = -16384
+
+        # Block 2L
+        # INVQAL
+        wd2 = QM4[wd1]
+        dlowt = (self._band[0].det * wd2) >> 15
+
+        # Block 3L
+        # LOGSCL
+        wd2 = RL42[wd1]
+        wd1 = (self._band[0].nb * 127) >> 7
+        wd1 += WL[wd2]
+
+        if wd1 < 0:
+            wd1 = 0
+        elif wd1 > 18432:
+            wd1 = 18432
+
+        self._band[0].nb = wd1
+
+        # SCALEL
+        wd1 = (self._band[0].nb >> 6) & 31
+        wd2 = 8 - (self._band[0].nb >> 11)
+
+        if wd2 < 0:
+            wd3 = ILB[wd1] << -wd2
+        else:
+            wd3 = ILB[wd1] >> wd2
+
+        self._band[0].det = wd3 << 2
+
+        # Block 4L
+        self._band[0].block4(dlowt)
+
+        return rlow
+
+    def higher_sub_band_decoder(self, higher_bits) -> int:
+        """Higher sub-band decoder for first two bits."""
+
+        # Block 2H
+        # INVQAH
+        wd2 = QM2[higher_bits]
+        dhigh = (self._band[1].det * wd2) >> 15
+
+        # Block 5H
+        # RECONS
+        rhigh = dhigh + self._band[1].s
+
+        # Block 6H
+        # LIMIT
+        if rhigh > 16383:
+            rhigh = 16383
+        elif rhigh < -16384:
+            rhigh = -16384
+
+        # Block 3H
+        # LOGSCH
+        wd2 = RH2[higher_bits]
+        wd1 = (self._band[1].nb * 127) >> 7
+        wd1 += WH[wd2]
+
+        if wd1 < 0:
+            wd1 = 0
+        elif wd1 > 22528:
+            wd1 = 22528
+        self._band[1].nb = wd1
+
+        # SCALEH
+        wd1 = (self._band[1].nb >> 6) & 31
+        wd2 = 10 - (self._band[1].nb >> 11)
+
+        if wd2 < 0:
+            wd3 = ILB[wd1] << -wd2
+        else:
+            wd3 = ILB[wd1] >> wd2
+        self._band[1].det = wd3 << 2
+
+        # Block 4H
+        self._band[1].block4(dhigh)
+
+        return rhigh
+
+
+# -----------------------------------------------------------------------------
+class Band(object):
+    """Structure for G722 decode proccessing."""
+
+    s: int = 0
+    nb: int = 0
+    det: int = 0
+
+    def __init__(self):
+        self._sp = 0
+        self._sz = 0
+        self._r = [0] * 3
+        self._a = [0] * 3
+        self._ap = [0] * 3
+        self._p = [0] * 3
+        self._d = [0] * 7
+        self._b = [0] * 7
+        self._bp = [0] * 7
+        self._sg = [0] * 7
+
+    def saturate(self, amp: int) -> int:
+        if amp > 32767:
+            return 32767
+        elif amp < -32768:
+            return -32768
+        else:
+            return amp
+
+    def block4(self, d: int) -> None:
+        """Block4 for both lower and higher sub-band decoder."""
+        wd1 = 0
+        wd2 = 0
+        wd3 = 0
+
+        # RECONS
+        self._d[0] = d
+        self._r[0] = self.saturate(self.s + d)
+
+        # PARREC
+        self._p[0] = self.saturate(self._sz + d)
+
+        # UPPOL2
+        for i in range(3):
+            self._sg[i] = (self._p[i]) >> 15
+        wd1 = self.saturate((self._a[1]) << 2)
+        wd2 = -wd1 if self._sg[0] == self._sg[1] else wd1
+
+        if wd2 > 32767:
+            wd2 = 32767
+
+        wd3 = 128 if self._sg[0] == self._sg[2] else -128
+        wd3 += wd2 >> 7
+        wd3 += (self._a[2] * 32512) >> 15
+
+        if wd3 > 12288:
+            wd3 = 12288
+        elif wd3 < -12288:
+            wd3 = -12288
+        self._ap[2] = wd3
+
+        # UPPOL1
+        self._sg[0] = (self._p[0]) >> 15
+        self._sg[1] = (self._p[1]) >> 15
+        wd1 = 192 if self._sg[0] == self._sg[1] else -192
+        wd2 = (self._a[1] * 32640) >> 15
+
+        self._ap[1] = self.saturate(wd1 + wd2)
+        wd3 = self.saturate(15360 - self._ap[2])
+
+        if self._ap[1] > wd3:
+            self._ap[1] = wd3
+        elif self._ap[1] < -wd3:
+            self._ap[1] = -wd3
+
+        # UPZERO
+        wd1 = 0 if d == 0 else 128
+        self._sg[0] = d >> 15
+        for i in range(1, 7):
+            self._sg[i] = (self._d[i]) >> 15
+            wd2 = wd1 if self._sg[i] == self._sg[0] else -wd1
+            wd3 = (self._b[i] * 32640) >> 15
+            self._bp[i] = self.saturate(wd2 + wd3)
+
+        # DELAYA
+        for i in range(6, 0, -1):
+            self._d[i] = self._d[i - 1]
+            self._b[i] = self._bp[i]
+
+        for i in range(2, 0, -1):
+            self._r[i] = self._r[i - 1]
+            self._p[i] = self._p[i - 1]
+            self._a[i] = self._ap[i]
+
+        # FILTEP
+        self._sp = 0
+        for i in range(1, 3):
+            wd1 = self.saturate(self._r[i] + self._r[i])
+            self._sp += (self._a[i] * wd1) >> 15
+        self._sp = self.saturate(self._sp)
+
+        # FILTEZ
+        self._sz = 0
+        for i in range(6, 0, -1):
+            wd1 = self.saturate(self._d[i] + self._d[i])
+            self._sz += (self._b[i] * wd1) >> 15
+        self._sz = self.saturate(self._sz)
+
+        # PREDIC
+        self.s = self.saturate(self._sp + self._sz)

tests/decoder_test.py

@@ -0,0 +1,47 @@
+# Copyright 2023 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# -----------------------------------------------------------------------------
+# Imports
+# -----------------------------------------------------------------------------
+import hashlib
+import os
+from bumble.decoder import G722Decoder
+
+
+# -----------------------------------------------------------------------------
+def test_decode_file():
+    decoder = G722Decoder()
+    output_bytes = bytearray()
+
+    with open(
+        os.path.join(os.path.dirname(__file__), 'g722_sample.g722'), 'rb'
+    ) as file:
+        file_content = file.read()
+        frame_length = 80
+        data_length = int(len(file_content) / frame_length)
+
+        for i in range(0, data_length):
+            decoded_data = decoder.decode_frame(
+                file_content[i * frame_length : i * frame_length + frame_length]
+            )
+            output_bytes.extend(decoded_data)
+
+        result = hashlib.md5(output_bytes).hexdigest()
+        assert result == 'b58e0cdd012d12f5633fc796c3b0fbd4'
+
+
+# -----------------------------------------------------------------------------
+if __name__ == '__main__':
+    test_decode_file()