How Ecasound Pitch Shifts Without Changing Tempo

Pitch shifting without affecting playback tempo requires decoupling frequency modification from time duration, a challenge fundamentally solved by digital signal processing (DSP) algorithms. Ecasound, a command-line multitrack audio processing utility, achieves independent pitch manipulation by combining resampling with time-scale modification (TSM) techniques—often implemented through internal buffer management, Waveform Similarity Overlap-Add (WSOLA), or by hosting frequency-domain phase vocoder algorithms via LADSPA plugin architectures.

The Core Challenge: Pitch vs. Speed

In conventional analog and simple digital playback, altering pitch inherently alters speed, much like speeding up or slowing down a vinyl record. Doubling the playback sample rate raises the pitch by an octave but halves the track's duration. To shift pitch while keeping tempo intact, an audio engine must modify the spectral content of the signal while synthetically preserving the original length of the audio stream.

The Two-Stage Process

To implement tempo-neutral pitch shifting, Ecasound typically applies a complementary two-stage pipeline:

1. Time-Scale Modification (Time-Stretching): The audio signal is expanded or compressed in time without altering its frequency content. If the intended goal is to raise the pitch, the audio duration is mathematically stretched.
2. Resampling (Sample Rate Conversion): The stretched audio is then resampled back down to match the original sample rate and duration. Resampling shifts all frequencies upward, raising the pitch, while compressing the previously stretched timeline back to the original playback tempo.

This pipeline can also operate in reverse: resampling the audio first to alter the pitch, followed by time-stretching or compressing to restore the original track duration.

Underlying Algorithms

Ecasound achieves the time-stretching component through two primary computational methods, either internally or via its supported plugin infrastructure:

1. Time-Domain Techniques (WSOLA)

In time-domain processing, such as Waveform Similarity Overlap-Add (WSOLA):

• The audio is segmented into small overlapping frames (grains), usually between 10 and 50 milliseconds in length.
• Grains are duplicated or dropped to adjust overall duration.
• The algorithm calculates cross-correlation across adjacent segments to locate mathematically optimal alignment points where waveforms match in phase.
• Segments are crossfaded together smoothly to eliminate clicks, pops, and transient disruption while preserving the fundamental pitch.

2. Frequency-Domain Techniques (Phase Vocoder)

For higher-fidelity harmonic tracking, phase vocoder techniques are utilized:

• Short-Time Fourier Transform (STFT): The audio stream is broken down into overlapping windowed frames and transformed into frequency and phase information.
• Spectral Redistribution: The phase and magnitude of individual frequency bins are shifted along the spectrum to represent the target pitch.
• Phase Unwrapping and Synthesis: The phase relationships between consecutive frames are recalculated to maintain temporal coherence and avoid hollow "phasey" artifacts.
• Inverse STFT: The modified bins are converted back into the time domain using the Inverse Fast Fourier Transform (IFFT) and reconstituted via overlap-add.

Ecasound’s Implementation and Chain Architecture

Ecasound is designed as an efficient, modular real-time signal routing engine. Because real-time pitch shifting is computationally intensive, Ecasound leverages its real-time buffer management and plugin architecture:

• Buffer Pipelines: Ecasound processes audio using configurable buffers (-b:buffersize). Pitch-shifting routines operate across these local sliding buffers, ensuring continuous low-latency playback without audio dropouts.
• LADSPA Integration: While Ecasound includes native chain operators for fundamental DSP tasks, specialized non-destructive pitch shifting is frequently executed through external LADSPA (Linux Audio Developer's Simple Plugin API) plugins hosted directly in an Ecasound chainsetup. Libraries like SoundTouch or Rubber Band provide the underlying WSOLA or phase vocoder algorithms, which Ecasound seamlessly feeds with audio streams, processes in real time, and outputs at the original sample clock.