Improve Speech Clarity with FFmpeg Speechnormalizer

This article provides a practical guide on how to use the speechnormalizer filter in FFmpeg to enhance the clarity and intelligibility of spoken audio. You will learn what the filter does, its key parameters, and how to apply it to your audio and video files using simple command-line examples to achieve consistent, professional-grade speech levels.

The speechnormalizer filter in FFmpeg is an audio filter designed specifically to improve speech clarity by dynamically adjusting volume levels. Unlike standard normalization, which scales the entire audio file based on a single peak, the speech normalizer analyzes the audio in small frames and adjusts the gain dynamically. This ensures that quiet words are boosted and excessively loud words are attenuated, resulting in a balanced and highly legible voice track.

Basic Usage

To apply the speech normalizer with its default settings, use the following basic command:

ffmpeg -i input.mp4 -af "speechnormalizer" output.mp4

This command processes the audio stream of input.mp4 through the speechnormalizer filter and saves the result to output.mp4 while keeping the default parameters, which are optimized for general speech.

Key Parameters

You can fine-tune the filter’s behavior by adjusting its parameters. The most important options include:

• peak (or p): Sets the target peak amplitude. The default value is 0.95 (representing 95% of maximum digital volume).
• rms (or r): Sets the target Root Mean Square (RMS) level, which represents the perceived loudness of the speech. The default is 0.25.
• threshold (or t): Sets the noise gate threshold. Audio below this level is treated as silence/noise and is not boosted. The default is 0.01. Raising this value prevents the filter from amplifying background hiss or room noise during pauses in speech.
• f: Specifies the frame length in milliseconds for analyzing the audio. The default is 250 ms.

Advanced Examples

To normalize speech while actively suppressing background noise during pauses, increase the threshold value:

ffmpeg -i input.wav -af "speechnormalizer=peak=0.9:threshold=0.05" output.wav

If you have a highly dynamic voice recording where the speaker goes from whispering to shouting, you can lower the target rms to prevent clipping and set a conservative peak level:

ffmpeg -i interview.mp3 -af "speechnormalizer=peak=0.85:rms=0.2" normalized_interview.mp3

By adjusting these parameters, you can customize the speechnormalizer filter to suit podcasts, voiceovers, video meetings, and any other audio where speech clarity is paramount.