Praat Tutorial: Getting Started with the Interface (2025)

Digital audio converts continuous sound waves into discrete data points through sampling. The sampling rate (44.1 kHz, 48 kHz, 96 kHz, 192 kHz) determines the highest reproducible frequency (Nyquist frequency). Uncompressed formats like WAV and AIFF store data exactly as captured, while compressed formats like MP3 use psychoacoustic models to remove imperceptible data, reducing file sizes by up to 90%. MP3 bitrate (kbps) differs from uncompressed bit depth (bits per sample). Sample rate selection depends on target medium: 44.1 kHz for CD, 48 kHz for video, 96-192 kHz for high-resolution audio.

Sound is a longitudinal wave consisting of compressions and rarefactions that propagate through air. A speaker converts electrical signals into sound through a voice coil (electromagnet) suspended in a permanent magnetic field. Waveform characteristics include frequency (cycles per second, measured in Hertz), period (time for one complete cycle), and amplitude (measured from peak to zero point). The Nyquist sampling theorem states that the sampling rate must be at least twice the maximum frequency to accurately reproduce a signal. Common sample rates are 44.1 kHz (CD standard) and 48 kHz, which can represent frequencies up to 22.05 kHz and 24 kHz respectively. Quantization converts continuous analog values into discrete digital values, with bit depth determining amplitude resolution: 8 bits provides ~48 dB, 16 bits provides ~96 dB, and 24 bits provides ~144 dB. The CD standard uses 44.1 kHz and 16-bit depth with Pulse Code Modulation (PCM) format. Lossless compression (FLAC) reduces file size without data loss, while lossy compression (MP3) permanently removes data for greater compression.

Digital audio is created through sampling, where continuous sound waves are captured at regular intervals to create digital representations; the Nyquist theorem states that to accurately reproduce sound, the sampling rate must be at least twice the highest frequency (20,000 Hz for human hearing), requiring a minimum of 40,000 samples per second, with consumer audio typically using 44.1 kHz and video audio using 48 kHz. Bit depth determines the number of amplitude values per sample, affecting dynamic range and audio quality. Audio formats include mono (single channel), stereo (two channels), and surround sound (5.1, 7.1, and spatial audio like Dolby Atmos). File formats are categorized as lossy (MP3, AAC, OGG - smaller file size but reduced quality), lossless (FLAC, ALAC - larger but no quality loss), and uncompressed (WAV - highest quality).

Digital audio uses sampling rates (44.1 kHz for CD, 96 kHz for DVD) based on Nyquist Theorem (sampling rate must be twice the highest frequency). Higher rates capture more harmonics for better detail. MIDI records musical data, not audio, while DAWs record actual audio. File formats like WAV and MP3 store digital audio data. Understanding these fundamentals is essential for proper audio production and mastering.

Digital audio recording involves measuring an analog voltage signal at discrete intervals and storing those measurements as digital numbers. The Nyquist-Shannon Sampling Theorem, developed by Harry Nyquist in the 1930s, establishes that at least two samples per cycle are needed to accurately capture any frequency. Since humans hear up to approximately 20 kHz, the minimum sampling rate should be 40 kHz. However, practical implementation required adding extra bandwidth for filter roll-off, resulting in the 44.1 kHz standard used for CDs. This rate allows for a gentle filter slope that sufficiently attenuates frequencies above 20 kHz while preserving audio quality.

This video demonstrates how spectrograms make speech sounds visually observable, allowing listeners to move beyond linguistic perception and analyze sounds as pure acoustic signals. When people hear /s/ and /f/ sounds, their brains automatically interpret them as distinct phonemes, which hinders detailed acoustic analysis. The spectrogram provides a visual representation that temporarily disables this linguistic processing, enabling viewers to observe the physical differences between similar sounds. The excerpt shows real-time spectrograms of the sounds /s/, /f/, /a/, /i/, and /u/, revealing their unique frequency patterns over time. This technique helps linguists and learners identify subtle acoustic distinctions that are otherwise imperceptible by ear alone. The video credits ScienceMusic.org for providing a real-time spectrogram tool that users can interact with to see how their own voice produces these sounds. It encourages viewers to explore the tool independently to gain hands-on experience with speech visualization. The video is a short excerpt from Lingthusiasm Episode 64, which focuses on making speech visible through spectrographic analysis, and directs viewers to additional resources for audio-only versions and further reading.

A spectrogram is a three-dimensional visualization tool that displays speech signals with time on the x-axis, frequency on the y-axis, and darkness (or color) representing amplitude on a dB scale; periodic sounds like voiced speech appear as vertical striations, noise sounds appear as white or light gray areas, and transients appear as brief vertical stripes, with the trade-off between time and frequency resolution determined by the window size used in spectrogram creation.

A spectrogram is a visual representation of speech signals that displays frequency (vertical axis) and time (horizontal axis), where formants (F1, F2, F3, F4) are peaks of energy that reveal articulatory information: F1 indicates mouth openness (low when mouth is closed, high when open), while F2 indicates tongue advancement (high for front vowels, low for back vowels); vowels show periodic resonances with higher amplitude than consonants, while voiceless consonants appear as noise without formants.

This video demonstrates how to analyze a spectrogram by identifying key acoustic features: formants (dark horizontal bands) indicate vowel quality with F1 representing tongue height and F2 representing tongue advancement; 'crappy formants' (weak, poorly defined bands) indicate sonorant consonants like nasals and laterals; burst releases with energy centered around 2000 Hz indicate velar stops, while energy concentrated above 4000 Hz indicates alveolar fricatives; and even noisy energy distribution across frequencies suggests non-sibilant fricatives like [f] or [θ]. By systematically mapping these acoustic landmarks, one can identify individual speech sounds and reconstruct spoken words from spectrographic data.

A spectrogram visualizes sound by showing frequency (vertical axis) against time (horizontal axis), with color indicating loudness. Higher resolution captures frequencies beyond human hearing. Any sound can be converted to an image, and any image can be converted back to sound. Artists have used this technology to embed hidden messages in music, creating visual representations of audio that reveal information not audible to the human ear.

Phonetic transcription is studied for three main purposes: (1) Pronunciation - to learn how to correctly pronounce words, (2) Articulation - to understand how to form sounds properly, and (3) Intonation - to learn the rise and fall of voice in speech. It helps learners understand where to pause, how to stress certain words, and how to stretch or shorten certain sounds in English.

Phonetic transcription is necessary because English orthography poorly maps letters to sounds (e.g., 'EA' in 'threat' vs 'meat' makes different sounds). The International Phonetic Alphabet provides unique symbols for each sound that remain consistent across languages. This universal system allows anyone who knows the phonetic alphabet to pronounce any language accurately, solving the problem of inconsistent spelling-to-sound relationships in English.

Phonetics aims to make speech sounds visible and practical to write down and print. This involves developing systems to describe sounds empirically in scholarly settings, allowing researchers to record, analyze, and share precise information about how sounds are produced and perceived across different languages and speakers.

Transcription is the primary representation of speech in non-audio format, but it is not theory-neutral—every transcription decision is an active analysis dependent on the transcriber's theories about language and speech. This includes decisions about text layout and encoding nonverbal elements like pauses and overlaps. The presentation introduces visualization methods as a way to re-examine and reimagine transcription, moving beyond text to explore quantitative approaches to discourse analysis and language variation.