Skip to content

decode_benchmark

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

decode_benchmark

README.md

Decode Benchmark for ESP32

Measures decoding performance for all four supported audio formats (FLAC, MP3, Opus, WAV) using DecoderSource::play_buffer(). Decodes ~10-second audio clips embedded in flash and reports timing statistics. No audio output.

Features

  • Embedded test audio in all four formats (no filesystem or network required)
  • Per-codec timing with Real-Time Factor (RTF)
  • Runs in a continuous loop for repeated measurements
  • Pre-configured for maximum performance (240 MHz, PSRAM, -O2)

Audio Source

Audio source: Beethoven Symphony No. 3 "Eroica", Op. 55, Movement I, Czech National Symphony Orchestra, from the Musopen Collection (public domain). 10-second excerpt starting at 1:00, downmixed to 48 kHz mono.

The same source recording is used by the microFLAC, microMP3, and microOpus decode benchmarks.

Format Bitrate File Size
FLAC lossless ~779 KB
MP3 128 kbps ~157 KB
Opus 64 kbps ~84 KB
WAV PCM 16-bit ~938 KB

Building and Running

PlatformIO

# Build for ESP32-S3
pio run -e esp32s3

# Build and flash for ESP32-S3
pio run -e esp32s3 -t upload -t monitor

# Build for ESP32
pio run -e esp32

ESP-IDF

idf.py set-target esp32s3
idf.py build flash monitor

Expected Output

ESP32-S3 @ 240 MHz:

=== microDecoder Decode Benchmark ===
Decoding ~10s clips, 48 kHz mono, no audio output

--- Iteration 1 ---
  Codec   File size     Decode time  Duration   RTF
  FLAC    798147 bytes     341.0 ms  10.00 s audio  RTF 0.0341
  MP3     161256 bytes     479.4 ms  10.06 s audio  RTF 0.0477
  OPUS     85663 bytes     799.6 ms  10.00 s audio  RTF 0.0800
  WAV     960264 bytes      69.1 ms  10.00 s audio  RTF 0.0069

RTF (Real-Time Factor) = decode_time / audio_duration. Lower is better:

  • RTF < 1.0: decoding is faster than real-time
  • RTF = 1.0: decoding keeps up with real-time playback
  • RTF > 1.0: decoding is slower than real-time

Regenerating Test Audio

To regenerate the embedded audio headers from a different source file:

  1. Download the source FLAC from Musopen or Archive.org
  2. Run the conversion script:
python3 convert_audio.py -i path/to/eroica.flac -s 60 -d 10

This extracts a 10-second clip starting at 1:00, encodes to all four formats (48 kHz mono), and generates C headers in src/.

File Structure

decode_benchmark/
├── src/
│   ├── CMakeLists.txt          # ESP-IDF component file
│   ├── main.cpp                # Benchmark code
│   ├── test_audio_flac.h       # Embedded FLAC data
│   ├── test_audio_mp3.h        # Embedded MP3 data
│   ├── test_audio_opus.h       # Embedded Opus data
│   └── test_audio_wav.h        # Embedded WAV data
├── CMakeLists.txt              # ESP-IDF project file
├── convert_audio.py            # Audio-to-header conversion script
├── partitions_4mb.csv          # Custom partition table
├── platformio.ini              # PlatformIO configuration
├── sdkconfig.defaults          # Shared ESP-IDF configuration
├── sdkconfig.defaults.esp32    # ESP32 PSRAM and flash settings
├── sdkconfig.defaults.esp32s3  # ESP32-S3 PSRAM, flash, and cache settings
└── README.md