Concurrent real-time interface for multimedia engines

The crime crate is useful for real-time multimedia applications for providing different output audio formats where the input is a stream of samples, typically coming from a machine-learning model which generates a fixed amount of samples in fp32.

This crate operates entirely on streams, provides delay compensation for each supported operation, and thus eliminates the need for providing padding and/or flushing the buffer. The first audio signal in the input is guaranteed to correspond to the first audio signal in the output, and likewise for the last signal.

Current functionalities:

• Input and outputs in futures::Stream.
• Resampling.
• Streaming audio codec (PCM, WAV, MP3, Ogg/Opus, WebM/Opus).
• Time scaling (using WSOLA).

Feature flags

Each audio format is behind an optional feature flag. No formats are enabled by default.

Feature	Format	Extra dependencies
pcm	Headerless PCM (linear + G.711 μ-law)	audio-codec-algorithms, half
wav	WAV (linear PCM with header)	half
mp3	MP3	mp3lame-encoder
ogg	Ogg/Opus	opus, ogg, crc
webm	WebM/Opus	opus

Example — enable MP3 and WAV only:

[dependencies]
crime = { version = "...", features = ["mp3", "wav"] }

Supported codecs

• pcm — Headerless sample stream
  • Linear PCM (i16, f16, f32, all little-endian)
  • G.711 μ-law
• wav — WAV
  • Linear PCM (i16, f16, f32, all little-endian)
• mp3 — MP3
• ogg — Ogg/Opus
• webm — WebM/Opus