SSRC - An audiophile-grade sample rate converter

You can watch introduction video at https://youtu.be/AlRDaNACx3Y.

You can download Windows binaries from Releases (under Assets).

If you have any thoughts or comments about this project, feel free to post them in Discussions.

Shibatch Sample Rate Converter (SSRC) is a fast and high-quality sample rate converter for PCM WAV files. It is designed to efficiently handle the conversion between commonly used sampling rates such as 44.1kHz and 48kHz while ensuring minimal sound quality degradation.

Features

• High-Quality Conversion: Achieves excellent audio quality with minimal artifacts.
• FFT-Based Algorithm: Utilizes a unique FFT-based algorithm for precise and efficient sample rate conversion.
• SleefDFT Integration: Leverages SleefDFT, a product of the SLEEF Project, for fast Fourier transforms (FFT), enabling high-speed conversions.
• SIMD Optimization: Takes advantage of SIMD (Single Instruction, Multiple Data) techniques for accelerated processing. It is capable of high-speed conversion using AVX-512.
• Dithering Functionality: Supports various dithering techniques, including noise shaping based on the absolute threshold of hearing (ATH) curve.
• Specialized Filters: Implements high-order filters to address the challenges of converting between 44.1kHz and 48kHz.
• Selectable Conversion Profile: You can select the filter lengths and computing precision. Single-precision computation is generally sufficient for audio processing, and even the standard profile allows for highly accurate conversion. However, since this tool is designed for audiophiles, you can also select a profile that performs all computation in double precision.
• Low-Latency Real-Time Processing: Suitable for demanding real-time applications by combining minimum-phase filters with an efficient partitioned convolution algorithm.

Why SSRC?

Sampling rates of 44.1kHz (used in CDs) and 48kHz (used in DVDs) are widely used, but their conversion ratio (147:160) requires highly sophisticated algorithms to maintain quality. SSRC addresses this challenge by using an FFT-based approach, coupled with SleefDFT and SIMD optimization, to achieve a balance between speed and audio fidelity.

See here for some experimental results.

Getting Started

Prerequisites

• Operating System: SSRC is compatible with multiple platforms.
• Audio Files: Input files should be in PCM WAV format.

Installation

1. Download the latest release from the GitHub repo.
2. Extract the downloaded archive to a directory of your choice.
3. Ensure the ssrc executable is accessible from your command line.

Usage

The basic command structure is as follows:

ssrc [options] <source_file.wav> <destination_file.wav>

You can also use standard input and output:

cat input.wav | ssrc --stdin [options] --stdout > output.wav

Options

| Option | Description | |----------------------------|------------------------------------------------------------------------------------------------| | --rate <sampling rate> | Specify the output sampling rate in Hz. Example: 48000. | | --att <attenuation> | Attenuate the output signal in decibels (dB). Default: 0. | | --bits <number of bits> | Specify the output quantization bit depth. Common values are 16, 24, 32. Use -32 or -64 for 32-bit or 64-bit IEEE floating-point output. Default: 16. | | --dither <type> | Select a dithering/noise shaping algorithm by ID. Use --dither help to see all available types for different sample rates. | | --mixChannels <matrix> | Mix, re-route, or change the number of channels. See the "Channel Mixing" section below for details and examples. | | --pdf <type> [<amp>] | Select a Probability Distribution Function (PDF) for dithering. 0: Rectangular, 1: Triangular. Default: 0. | | --profile <name> | Select a conversion quality/speed profile. Use --profile help for details. Default: standard. | | --minPhase | Use minimum-phase filters instead of the default linear-phase filters, which makes the processing delay negligible. | | --partConv <log2len> | Divide a long filter into smaller sub-filters so that they can be applied without significant processing delays. | | --st | Disable multithreading (enabled by default). | | --dstContainer <name> | Specify the output file container type (riff, w64, rf64, etc.). Use --dstContainer help for options. Defaults to the source container or riff. | | --genImpulse ... | For testing. Generate an impulse signal instead of reading a file. | | --genSweep ... | For testing. Generate a sweep signal instead of reading a file. | | --stdin | Read audio data from standard input. | | --stdout | Write audio data to standard output. | | --quiet | Suppress informational messages. | | --debug | Print detailed debugging information during processing. | | --seed <number> | Set the random seed for dithering to ensure reproducible results. |

Example

Convert a WAV file from 44.1kHz to 48kHz with dithering:

ssrc --rate 48000 --dither 0 input.wav output.wav

Conversion Profiles

Profiles allow you to balance between conversion speed and quality (stop-band attenuation and filter length).

| Profile Name | FFT Length | Attenuation | Precision | Use Case | |--------------|------------|-------------|-----------|----------------------------------------| | insane | 262144 | 200 dB | double | Highest possible quality, very slow. | | high | 65536 | 170 dB | double | Excellent quality for audiophiles. | | long | 32768 | 145 dB | double | Superb quality. | | standard | 16384 | 145 dB | single | Great quality, default setting. | | short | 4096 | 96 dB | single | Good quality. | | fast | 1024 | 96 dB | single | Good quality, suitable for most uses. | | lightning | 256 | 96 dB | single | Low latency, suitable for real-time uses. |

You can see all profiles and their technical details by running ssrc --profile help.

Channel Mixing (--mixChannels)

The --mixChannels option allows you to mix, re-route, or change the number of channels using a matrix string.

• Syntax: The matrix string is a series of numbers separated by commas (,) and semicolons (;).
  • Commas (,) separate the gain values for each column in a row.
  • Semicolons (;) separate the rows.
• Logic:
  • The number of rows in the matrix defines the number of output channels.
  • The number of columns in the matrix must match the number of input channels.

Example 1: Stereo to Mono Downmix

To combine a 2-channel stereo input into a 1-channel mono output, you can use a 1-row, 2-column matrix. The standard formula is Mono = 0.5 * Left + 0.5 * Right.

--mixChannels '0.5,0.5'

Example 2: Mono to Stereo

To duplicate a 1-channel mono input into a 2-channel stereo output, you can use a 2-row, 1-column matrix.

--mixChannels '1;1'

This sets both the left and right output channels to be equal to the mono input channel.

Example 3: Swapping Stereo Channels

To swap the left and right channels of a stereo file, you need a 2x2 matrix. The goal is to make the new left channel equal to the old right channel, and the new right channel equal to the old left channel.

--mixChannels '0,1;1,0'

• The first row 0,1 means Output0 = (0 * Input0) + (1 * Input1).
• The second row 1,0 means Output1 = (1 * Input0) + (0 * Input1).

Spectrum Analyzer (scsa)

The project includes scsa, a command-line spectrum analyzer. While it can be used as a general-purpose analyzer, it is primarily designed for automated testing and verification, for example in a CI environment.

Purpose and Features

• Automated Testing: The primary purpose of scsa is to check audio spectra against predefined criteria, making it ideal for automated quality assurance in a CI/CD pipeline.
• Cross-Platform and Dependency-Free: As a command-line tool, it does not rely on any GUI libraries or have OS-specific dependencies, making it highly portable and easy to integrate into various workflows.
• SVG Output for Debugging: When a test fails, scsa can generate an SVG image of the spectrum. This visual output is extremely useful for identifying the cause of the failure. An SVG is also generated if no check file is provided, allowing scsa to be used as a general-purpose analyzer.
• High-Precision Analysis: Unlike many standard analyzers, all internal processing is performed in double precision. This minimizes the impact of floating-point noise, allowing for highly accura