Merge pull request #3 from KrisKennaway/audio

Improvements to audio processing (quality, support better quality video stream on //gs)

.gitignore

@@ -1,3 +1,4 @@
 __pycache__/
 venv/
 videos/
+*.a2m

player/main.s

@@ -437,10 +437,10 @@ recv: ; 15 cycles so far
 ; pad cycles to keep ticking on 36/37 cycle cadence
 ; TODO: what can we do with the luxury of 14 unused cycles?!
 @2: ; 30 so far
+    STA TICK ; 4 ; 34
     ; X will usually already be 0 from op_ack except during first frame when reading
     ; header but reset it unconditionally
     LDX #$00 ; 2
-    STA TICK ; 4 ; 36
 
     NOP ; 2
     STA dummy ; 4
@@ -453,7 +453,7 @@ op_nop:
     LDY WDATA ; 4
     STY @D+1 ; 4
 @D:
-    JMP op_nop ; 3 ; 23 with following tick (37 in fallthrough case)
+    JMP op_nop ; 3 ; 23 with following tick (39 if we fell through from checkrecv case)
 
 ; Build macros for "fat" opcodes that do the following:
 ; - tick twice, N cycles apart (N = 4 .. 66 in steps of 2)
@@ -1312,16 +1312,18 @@ op_ack:
     LDA #>S0RXRD ; 2
     STA WADRH ; 4
     LDX #<S0RXRD ; 2
-    STX WADRL ; 4
+    ; prepare for ADC below, but reordered to allow TICK at offset 34
+    CLC ; 2
 
-    STA TICK ; 4 (36)
+    STA TICK ; 4 (34)
+
+    STX WADRL ; 4
 
     LDA WDATA ; 4 Read high byte
     ; No need to read low byte since it's guaranteed to be 0 since we're at the end of a 2K frame.
 
     ; Update new Received Read pointer
     ; We have received an additional 2KB
-    CLC ; 2
     ADC #$08 ; 2
 
     STX WADRL ; 4 Reset address pointer, X still has #<S0RXRD
@@ -1339,7 +1341,7 @@ op_ack:
     ; - used as the low byte for resetting the W5100 address pointer when we're ready to start processing more data
     LDX #$00 ; 2 restore invariant for dispatch loop
 
-    JMP checkrecv ; 3 (37 with following STA TICK)
+    JMP checkrecv ; 3 (39 with following STA TICK)
 
 ; Quit to ProDOS
 exit:

transcoder/audio.py

@@ -8,21 +8,38 @@ import numpy as np
 
 
 class Audio:
+    """
+    Decodes audio stream from input file and resamples.
+
+    Notes on audio bitrate:
+
+    At 73 cycles/tick, true audio playback sample rate is
+    roughly 1024*1024/73 = 14364 Hz (ignoring ACK slow path).
+    Typical audio encoding is 44100Hz which is close to 14700*3
+    Downscaling by 3x gives better results than trying to resample
+    to a non-divisor.  So we cheat a bit and play back the video a
+    tiny bit (<2%) faster.
+
+    For //gs playback at 2.8MHz, the effective speed increase is only about
+    1.6x.  This is probably because accessing the I/O page is done at 1MHz
+    to not mess up hardware timings.
+
+    This is close (2.1%) to 22500Hz which is again a simple divisor of the
+    base frequency (1/2).
+    """
+
     def __init__(
-            self, filename: str, normalization: float = None):
+            self,
+            filename: str,
+            bitrate: int = 14700,
+            normalization: float = None):
         self.filename = filename  # type: str
 
         # TODO: take into account that the available range is slightly offset
         # as fraction of total cycle count?
         self._tick_range = [4, 66]
 
-        # At 73 cycles/tick, true audio playback sample rate is
-        # roughly 1024*1024/73 = 14364 Hz (ignoring ACK slow path).
-        # Typical audio encoding is 44100Hz which is close to 14700*3
-        # Downscaling by 3x gives better results than trying to resample
-        # to a non-divisor.  So we cheat a bit and play back the video a tiny
-        # bit (<2%) faster.
-        self.sample_rate = 14700.  # type: float
+        self.sample_rate = float(bitrate)  # type: float
 
         self.normalization = (
                 normalization or self._normalization())  # type: float
@@ -39,7 +56,8 @@ class Audio:
 
         a = librosa.core.to_mono(data)
         a = librosa.resample(a, f.samplerate,
-                             self.sample_rate).flatten()
+                             self.sample_rate,
+                             res_type='scipy', scale=True).flatten()
 
         return a
 

transcoder/main.py

@@ -20,6 +20,11 @@ parser.add_argument(
     '--audio_normalization', type=float, default=None,
     help='Override auto-detected multiplier for audio normalization.'
 )
+parser.add_argument(
+    '--audio_bitrate', type=int, default=14700,
+    help='Select output audio bitrate (Hz), controls video speed (Default: '
+         '14700; try 22500 for //gs 2.8MHz mode)'
+)
 parser.add_argument(
     '--every_n_video_frames', type=int, default=2,
     help='Allows skipping frames of input video to lower effective output '
@@ -42,6 +47,7 @@ def main(args):
     m = movie.Movie(
         filename,
         every_n_video_frames=args.every_n_video_frames,
+        audio_bitrate=args.audio_bitrate,
         audio_normalization=args.audio_normalization,
         max_bytes_out=1024. * 1024 * args.max_output_mb,
         video_mode=video_mode.VideoMode[args.video_mode],

transcoder/movie.py

@@ -15,6 +15,7 @@ class Movie:
     def __init__(
             self, filename: str,
             every_n_video_frames: int = 1,
+            audio_bitrate: int = 14700,
             audio_normalization: float = None,
             max_bytes_out: int = None,
             video_mode: VideoMode = VideoMode.HGR,
@@ -27,7 +28,8 @@ class Movie:
         self.palette = palette  # type: Palette
 
         self.audio = audio.Audio(
-            filename, normalization=audio_normalization)  # type: audio.Audio
+            filename, bitrate=audio_bitrate,
+            normalization=audio_normalization)  # type: audio.Audio
 
         self.frame_grabber = frame_grabber.FileFrameGrabber(
             filename, mode=video_mode, palette=self.palette)