Contents

What is Signalyze?

Signalyze is an interactive program for the analysis of speech and other acoustic material. It contains a large set of signal editing, signal analysis and signal manipulation tools. Signalyze runs only on Macintosh computers . Buy Signalyze and a PowerMac, G3 or G4, and you have all you need to record, analyze and reproduce professional 16-bit sound.

Uses for Signalyze

Stimuli for perception experiments

• Splice two signals together
• Mix noise into a speech signal
• Align dichotic stimuli

Interactive speech analysis

• Make superb 256-grayscale or color spectrograms
• Measure duration, frequency and amplitude
• Perform pitch extractions
• Slow down or speed up speech
• Filter signals
• Manipulate signals
• Change the sampling frequency of a signal

Foreign Language Teaching

• Work on an AV- or Power Macintosh or plug in an Apple®, MacRecorder®, MacAdiosª or AudioMedia digitizer
• Record a sentence,
• Replay the recording,
• Show its fundamental frequency (intonation pattern)

Signalyze Overview

The following figure presents an overview of Signalyze's 3 main windows. The Signalyze Test signal is loaded into the first signal track, with labels showing and a portion selected. In the Information Bar can be seen the frequency equivalent ("f 2.56") of the selected duration ("d 391.19"), beginning at the left edge of the selection (time: "ms 871.6") (units are user-specified). A wide-band spectrogram of the whole signal is displayed in the second signal track.

Signalyze Capabilities

Signalyze is multi-lingual

The program switches instantaneously into English, French or German. Menus, buttons, information dialogs, the Onboard Help, and Balloon help switch instantly into the desired language. Onboard Help is also available in Italian.

Up to 100 simultaneous signals

How? By showing the signals on cards. Each card can show as many signals as fit on a screen (or fewer). To see more signals, you simply jump to another card.

All 100 signals can be synchronous

Move one signal, and all other signals stay time-aligned (even signals on other cards). Or keep any signal stationary while moving any other signals.

Signalyze is Fast

• The programÕs written in tight C-code, with extensive use of pointer logic.
• Wherever possible, fast integer math is favored over slower floating-point math
• Some floating-point routines take advantage of a co-processor, if available
• Time-critical routines are optimized through the use of register variables

Signalyze in the future will be rewritten in PPC ÒnativeÓ code for faster operations. But Signalyze will stay a ÒuniversalÓ application; a 68k version for older machines will also be kept updated as necessary.

8- or 16-bit sampling

Signalyze by default operates on 16-bit signals for best sound quality:

¥ 16-bit signals are less difficult to record. With 16-bit signals, thereÕs less danger of excessive or insufficient recording amplitude. Record spontaneous speech, childrenÕs or pathological speech with much less fuss than with 8-bit signals. ¥ They encode high-frequency information more reliably. Frequency information above 2 kHz is weak or insufficient in 8-bit signals. 16-bit signals give you the full range. ¥ They encode the full range of amplitude information. 8-bit signals only give you 42 dB of amplitude. 16-bit signals can encode about 90 dB — which covers the usual range of speech, music and animal sounds.

Signalyze handles 8-bit signals just as well, if desired. They are automatically translated into 16-bit format for most operations, and can be saved in either format. (For much of speech analysis—especially for time measures and stimuli presentation—8-bit signals are just fine.)

Signalyze is User-Friendly

Signalyze Tools

Signal Editing Tools

	Copy, cut, paste and clear whole signals or parts of signals
	Optional splicing at signal junctions
	Optional cosine taper at signal ends

Slow down or Speed up Speech

• Slows speech down by up to five times (500%)
• Speeds speech up by up to five times (20%)
• At 75-150%, sound quality is largely preserved for most voices
• Optionally save modified signals

Easy manual scoring

ÒManual scoringÓ means getting numeric values directly from the signal. In this way, you can easily obtain—

• the duration of a speech sound
• the frequency from a spectral peak
• amplitude differences between two vowels

and much moreÉ

Get Duration

1. Blacken part of a signal
2. Click on the duration field
3. The duration is shown in the Edit Window
4. Double-click or press Return to transfer duration to the Text Window

Get Frequency

1. Place the cursor at the desired location in the signal.
2. Click on spectrum button.
   A spectrum is produced.
3. Click in the spectrum while holding down the SHIFT and COMMAND keys.
   The frequency and original time in the raw signal are displayed.

Get Amplitude Differences Between Two Vowels

1. Make power or RMS envelope from signal 2. Blacken space between two peaks 3. Request amplitude difference through statistics command 4. dB amplitude difference is shown in edit window

Get Amplitude or Frequency Differences from Spectra

1. Click in signal 2. Make spectrum 3. Blacken distance between two peaks

An Extensive labeling facility

• Easy to use
• Coded for up to nine levels (e.g., syllable, segment, etc.)
• Label points or sections of the signal
• Saved as TAB-delimited text files
• Use phonetic font or any other font & font size
• Edit existing labels
• Insert or add new labels anywhere
• Delete any label or all labels
• Search by label or level
• Two label search patterns; switch between search patterns with the shift lock key

Spectral Analysis Tools

Signalyze provides a wide range of tools suitable for the spectral analysis of speech:

• Narrow-band and wide-band spectra (seven spectral bandwidths: 300 Hz, 200 Hz, 125 Hz, 40 Hz, 33 Hz, 20 Hz, 10 Hz)
• Narrow-band and wide-band spectrograms at the named bandwidths
• Spectral zoom (full/half/quarter range)
• High-quality spectrogram displays for grayscale and color (16 colors/grays and 256 colors/grays)
• Autocorrelation (inverse) filtering for Fo
• Cepstra and cepstrograms
• Linear Predictive Coding (LPC), orders 8-25
• Cone kernel spectra and spectrograms (256/512 pts.)
• Long-term averaged spectrum

Some examples

Narrow-band spectrum of a female speaker (Fo = 305 Hz, H2 = 605 Hz)

Wide-band spectrum of an /a/-vowel Formant 1 = 580 Hz, Formant 2 = 1220 Hz

LPC spectrum of /æ/-vowel Formant 1 = 700 Hz Formant 2 = 1350 Hz

Narrow-band cepstrum of female voice (Fo cycle duration = 3.5 ms, Fo = 286 Hz)

Narrow-band spectrogram of syllable with falling intonation (produced as grayscale spectrogram in Signalyze and dithered with the freeware program NIH Image)

Wide-band spectrogram of the syllable /ka/ Produced as grayscale spectrogram in Signalyze and dithered with the freeware program NIH Image. You need a 256 grayscale/color monitor, but you can print this on any printer.

Spectrogram with scales Old-style black-and-white Macintosh dithering, as it comes out of Signalyze on a black-and-white Macintosh, printed on any printer. Please note that the scales are given in round numbers.

Wide-band spectrogram of the syllable /ka/ 256-grayscale spectrogram. You need a grayscale or color monitor to see this image correctly. To print this out, you need a grayscale or color-capable printer.

Waterfall representation (Obtained with the freeware program NIH Image from the same spectrogram as used in the previous figures. Side view.)

Waterfall representation (Obtained with the freeware program NIH Image from the same spectrogram as used in the previous figures. Front view.)

Wide-band FFT (left) and Wide-band Cone Kernel (right) side by side. Cone Kernel techniques provide much greater precision in the temporal and frequency domains. Particularly useful for formant determination. Printed with the color/grayscale option.

A signal waveform and an LPC-gram of the word ÒSignalyzeÓ. When well adjusted, LPCÕs show rather clear formant tracks. Printed with the color/grayscale option.

Fundamental frequency

Signalyze provides three algorithms for Fo extraction:

• Temporal structure analysis
• FFT Comb
• Autocorrelation

Why three algorithms? Because signals are not all alike. Each algorithm has its advantages and disadvantages. Signalyze lets you choose the algorithm that fits your data best.

Limits of Fundamental Frequency Analysis
Fundamental frequency extraction is rarely an easy task. When Signalyze was developed, algorithms existed that provide a very high reliability in extracting fundamental frequency from a signal, but at the time these were too processor-intensive for an interactive micro-computer application like Signalyze. The algorithms incorporated in Signalyze aim for a reasonable compromise between speed and reliability.

Some Examples

Temporal Structure Analysis (TSA)

Advantages Fast Continuous frequency response Ignores information in harmonics One extraction every pitch period Disadvantages Frequency window must be defined closely Has difficulty with frequency shifts much greater than an octave Male speaker, /ka/, falling intonation

FFT-Comb Analysis

Advantages Quite robust One frequency window fits all voices Uses information from harmonics Disadvantages Quantized frequency steps Cannot handle divergences between fundamental frequency and harmonics Male speaker, /ka/, falling intonation, output filtered to reduce effect of quantization

Autocorrelation

Advantages Quite robust One frequency window fits all voices Uses information from harmonics Disadvantages Quantized frequency steps Cannot handle divergences between fundamental frequency and harmonics Male speaker, /ka/, falling intonation, output filtered to reduce effect of quantization

Filtering

Signalyze provides a number of software filters. Most prominent are the well-known Butterworth and Chebychev filters. There are also very fast 3-point triangular and 7-point hamming filters, as well as a simple moving average (convolution) filter.

Some examples

Butterworth Filter Top: Signal of /a/-vowel Middle: Narrow-band spectrum without filtering Bottom: Narrow-band spectrum after high-pass filtering at 800 Hz (5th order Butterworth)

Chebychev Filter Top: Signal of /a/-vowel Middle: Narrow-band spectrum without filtering Bottom: Narrow-band spectrum after high-pass filtering at 800 Hz (5th order Chebychev with 10 dB ripple)

3-point Triangular Filter Top: Signal of male speaker pronouncing /m'orgen/ Middle: Narrow-band spectrum without filtering Bottom: Narrow-band spectrum after application of triangular filter (select ÒsmoothingÓ in spectral analysis setup)

Signal Modification

Signalyze provides numerous routines to modify signals:

• arithmetic transformations
• transcendental transformations
• slow down or speed up speech signals
• down- and up-sampling (from and to any frequency 50-100,000 Hz)
• derivative differences
• power and RMS envelopes
• splines
• zero-crossings
• trim off quiet signal ends that waste disk space

Some examples

Down-sampling Substitute an average for the first of every two samples and delete the second sample.

Up-sampling Add interpolated samples between pairs of samples.

Differentiation Calculate differences between succeeding samples.

Splines Calculate spline curves and their derivative signals.

Sound Input and Output

Sound input and output is easy with Signalyzeª running on the Mac.

Old-style Input and Output

Input: Prior to System 7.0, each device had to be supported individually. Signalyze supports three such devices:

• MacroMindÕs MacRecorder (anti-aliased):

• Works with all Macs, incl. PowerBooks
• Optional recording directly to disk

• DigidesignÕs AudioMedia, Audiomedia II and AD IN.
• Some GW Instruments AD boards and the Model 411. Be sure to get an anti-alias filter.

Output: Signalyze has suitable playback via the built-in loudspeaker on old-style Macintoshes (Mac Plus, Mac SE, Portable, etc.)

New-style Input and Output

Starting with System 7.0, the Macintosh supports sound input and output directly via the Sound control device (found in the Control Panel in the Apple Menu). Signalyze supports all devices defined in this way. This includes:

• AppleÕs built-in microphone (Macintosh LC, IIsi, IIci, PowerBooks, etc.)
• AppleÕs built-in loudspeaker (any Macintosh)
• Any other device for which a System 7.0-compatible driver exists. This includes AV-Macintoshes and Power Macintoshes, at the maximum bitwidth that the machine supports (AV-Macintoshes and Power Macintoshes: 16 bits. Other machines: 8 bits). For best results, on pre-Power Macintosh computers you should make sure that you have Sound Manager 3.0 (available from Apple) in your System Folder. The Power Macintosh includes Sound Manager 3.0 in its system (OS 8 & 9). This driver assures the best sound input and output quality on all Macintoshes.

With Signalyze running on a Power Macintosh, you have all you need for professional speech signal analysis.

File Input and Output

Signalyze can read and write most common Macintosh file formats for signals. Additional signal formats are available through the freeware program FileConverter (more information is below). Signalyze also supports common text and graphics formats.

Signal File Formats

Signalyze can read and write signals in the following formats:

• Signalyze
• AIFF mono/stereo
• AudioMedia I & II/Sound Designer II mono/stereo
• MacADIOS
• SoundEdit/MacRecorder mono/stereo
• A simplified SoundEdit Pro format
• SoundWave
• Common sound (snd) resource formats
• ASCII-text formats (single channel)

You can simultaneously load as many files as will fit into the program (batch load). Automatic up/down-sampling is performed when saving to non-Signalyze, fixed sampling-frequency formats.

FileConverter

The FileConverter utility converts between Signalyze and other formats, including:

• Digidesign: AudioMedia, Sound Designer I/II mono/stereo
• AIFF, mono/stereo
• MacSpeech Lab/MacADIOS
• SoundEdit/MacRecorder, mono/stereo
• Sound Wave/Impulse
• Engineering Design
• ADF (BLISS, CSRE) format
• Europec-SAM (European sound audio format)
• Haskins Lab PCM format
• NSP (Kay)
• Phonetics-Munich format
• Sun/NeXT µ-law
• Generic binary (Intel-type and Motorola-type)

FileConverter is available for download via the downloads page. FileConverter can convert all files in a given folder at once, and can work in background mode.

Text Format

Numeric results are stored in a text editor. This information can be saved and re-opened in TEXT format (compatible with common text processing, spreadsheet and statistics programs: Excel, StatView, SYSTAT, etc.).

Graphics Format

Spectrograms and other graphs can be saved and re-opened in PICT format (compatible with most graphics programs: Microsoft Word, Canvas, Claris Works, SuperPaint, PhotoShop, NIH Image, etc.).

The Documentation

The documentation is packed with information:

• A quick guide to Version 3.0 additions.
• Signalyze Guided Tour: Shows you how to do (nearly) everything you can do with Signalyze.
• Numerous Òhow-toÓ summaries for common speech analysis techniques.
• Installation suggestions.
• Signalyze UserÕs Reference: Keys, buttons, menus, setup dialogs in detail.
• Signalyze Technical information: Ergonomics, memory management, file formats, numeric formats.

It is one of the most extensive and most detailed documentation that comes with any speech analysis system.

Computer Requirements

Signalyzeª runs on any Macintosh starting with the Macintosh Plus. The program takes advantage of larger screens by showing more signals per screen. Much improved spectrograms are possible on Macs equipped with grayscale and color monitors. The program is System 7.0-compatible, and can run in the background with adjustable background priority.

Some Common Configurations

Macintosh Classic, Macintosh SE or Macintosh Plus

This is a setup which gives ÒbasicÓ service. Main complaints: Screen too small, response sluggish. Recommendation: Get the full 4 Mb and operate in MultiFinder or System 7.0. Better yet, upgrade your systemÉ

Macintosh LC or Classic II

The LC is a step up from the Macintosh Classic series because you get a larger grayscale or color screen and a faster computer. The Classic II gives good speed but has insufficient screen space. DonÕt expect too much performance from the built-in mic. MacRecorder is better.

Macintosh II series (LC II, II, IIcx, IIci, IIsi, IIfx, etc.)

This is an adequate environment to run Signalyze. Recommendations: Get a minimum of 6 Mb of memory and an Apple-compatible Laser printer.

Quadras and Centris

Compared with the older Macs above, the speed of the machine (with all caches enabled) is something to see. There are no known incompatibilities with these machines.

PowerBooks and Portable Mac

Signalyze runs just fine on the whole series of Macs-on-the-Go, including the latest G3 PBÕs. Memory can get a bit tight at times on the old models that have not had their amount of RAM upgraded, but virtual memory (System 7 and up) or RAM Doubler from Connectix can help out there.

AV-Macintoshes and Power Macintoshes

Signalyze 3.12 remains workable on the latest generation of Macs. The outstanding issues caused by changes in the Macintosh hardware and OS are well-documented in the Known Issues document, and will be addressed in an upcoming update (free for registered version 3 users).

Because of additional image processing capacity, spectrograms are quite a bit faster. The built-in 16-bit sound input and output works well. You need no extra board on these machines, just an adequate amplifier or the AV Monitor. This is our current top recommendation for a university-based speech analysis station.

Memory Requirements

Signalyzeª Version 3 ships with its memory set to a recommended 2 megabytes. If youÕre running System 7 on an older machine, this means that you will likely need a minimum of 4M RAM total installed to support your System and Signalyze. But for added comfort, do consider getting more—even on older Macs, a minimum of 8M is recommended to help stabilize System 7.

If you are running OS 8 or 9 on a newer Mac, 32M is a recommended minimum; 64M or more is better. In this case you should be able to increase SignalyzeÕs memory up to as much as 16M to ensure plenty of recording time and calculation memory for handling complex signal analysis and editing.

The Cost?

Less than you think:

Individual license: $350.

Departmental license: $700.

The cost of the license includes full support from Linguist Plus and free upgrades to any version 3 revisions.

Note: A department is defined as a separate budget entity (department, laboratory, research group, etc.) within a larger organization. Departmental licenses are channeled via a single department administrator, who is responsible for ensuring the conditions of the Signalyze License Agreement are adhered to.

Support

Support is available for registered users of version 3 via several sources:

• The Signalyze Site contains a great deal of information on running Signalyze, and on maintaining your Mac workstation in general:

  http://www.signalyze.com

• The Signalyze discussion mailing list, SigList, has over 100 speech research scientists from around the world subscribed. If you need help or information from a colleague who also uses Signalyze, this is the place to find it.

  To subscribe to the SigList, please contact Linguist Plus with your registration information at support@signalyze.com.

• You can also get individual help directly from Linguist Plus, Inc. via email, in either English, Japanese or French (sorry, support in German is no longer available):

  support@signalyze.com

Please note that help in French will require a turnover of 2-3 days; requests for help in English or Japanese should be responded to within 24 hours.

Availability and Ordering Information

Signalyzeª is available now through its exclusive international distributor:

Purchase can be made by credit card via Agora Language MarketplaceÕs web site at the URL above. POÕs from educational institutions are also welcome. Please contact Agora for more information.

La version française

Ces dernières années, Signalyzeª a été bien accueilli dans les pays francophones, particulièrement en France et en Suisse. Par conséquent, nous avons fait un effort spécial pour convertir tous les aspects du logiciel en français. On vous offre le suivant:

• Une interface française. Au moyen dÕune simple sélection de menu, vous pouvez choisir lÕinterface française. Tous les menus, boîtes de dialogue, messages, etc., seront affichés en français.
• Aide en français. Une fois que vous fonctionnez en mode français et dans la mesure où le logiciel se trouve dans le même dossier où on retrouve le fichier «Aide-Signalyze», vous pouvez consulter Aide en français. Aide vous fournit des informations sur toutes les fonctions du logiciel ainsi que de précieux conseils concernant son utilisation générale. Il suffit de cliquer sur un des multiples boutons «Aide» quÕon retrouve un peu partout dans le logiciel.
• Les bulles dÕaide sous Système 7. Après avoir activé les bulles dÕaide, de petits messages en français vous guident dans lÕapprentissage des différentes composantes de lÕinterface du logiciel.
• Le manuel en français. Le manuel est dorénavant disponible en français (270 pages). Vous y trouverez une introduction détaillée à lÕemploi et au fonctionnement du logiciel.

Die deutschsprachige Version

Signalyzeª funktioniert zu großen Teilen auf Deutsch:

• Eine deutschsprachige Benutzeroberfläche. Mittels einer einfachen Menüeinstellung können Sie die Benutzeroberfläche auf Deutsch umstellen. Alle Menüs, Dialogboxen, Fehlermeldungen, usw. erscheinen auf Deutsch.
• ãHilfeÓ in Deutsch. Wenn Sie die deutschsprachige Oberfläche eingeschaltet haben, und wenn Signalyze sich im gleichen Ordner wie ãSignalyze-HilfeÓ befindet, erscheint ãHilfeÓ auf Deutsch. ãHilfeÓ gibt Erklärungen über die verschiedenen Funktionen sowie nützliche Hinweise zur allgemeinen Bedienung des Programms. Man braucht nur auf einen der ãHilfeÓ-Knöpfe zu klicken, die ein bisschen überall im Programm zu finden sind.
• Die ãHilfeÓ-Ballons unter System 7. Wenn Sie unter System 7 die ãHilfeÓ-Ballons eingeschaltet haben, erscheinen über der Signalyze Benutzeroberfläche hilfreiche Meldungen, die Ihnen das Erlernen der Oberfläche erleichtern.

Credits and Philosophy

Signalyzeª has been written by Eric Keller with assistance from (in alphabetical order): Philip Keller, Lorenzo Morellini, Pascal Perrier, Thomas Styger and Stefan Werner, the valued beta testers, and many others whose precious time and code found its way into the program.

Eric Keller (the author of Signalyze) holds a Ph.D. in Linguistics and is professor for computer science at the University of Lausanne, Switzerland. HeÕs published in the phonetic sciences for the last 18 years.

Signalyze was developed in private initiative as an extension of software used for internal purposes in the ÒLaboratoire dÕanalyse informatique de la paroleÓ of the University of Lausanne, Switzerland. The programÕs development was not supported by public funds. The University of Lausanne is not in any way responsible or legally liable for this product. The programÕs price structure is primarily geared to the recuperation of current costs and not to the generation of profit, nor to the recuperation of the programÕs real development costs.

The development of this program benefited from interaction with a large number of users and speech specialists. Over the years, many user comments translated into improvements and new features.

June 2001 update

Signalyze is now handled by Linguist Plus, Inc., a registered corporation of Washington State, U.S.A. Linguist Plus was founded by Kevin Varden, a linguist with a long-standing interest in computing. Kevin began using Signalyze 7 years ago while finishing his University of Washington dissertation, and has been helping support Signalyze for the past 6. He has worked at several universities in Japan, and is currently at the University of Tsukuba.

Updates and further development of Signalyze are planned by Linguist Plus, although no update release schedule has yet been fixed.

Support in English or Japanese (typical 24-hour turnover) and French (typical 2-3 day turn-over) is available through support@signalyze.com.

Trademarks

Signalyze and Linguist Plus are trademarks of Linguist Plus, Inc., a registered corporation of the State of Washington, U.S.A.

Other trademarks mentioned at this site are either trademarks or registered trademarks of their respective owners.

A Word of Caution

A tremendous amount of effort was expended to make Signalyze as accurate as possible and to ensure its proper functioning. However, given the programÕs great complexity, and the sweeping changes to the Macintosh architecture and OS since the last update, some minor compatibility issues have arisen between Signalyze and the PowerMac line. These and available work-arounds are well-documented in the Known Issues document.

Linguist Plus, Inc. will not be held responsible for any consequences arising from any errors of accuracy generated in the course of your research. It is your responsibility to check the accuracy of all results of the program.

In some countries and states, a producer may not refuse to assume responsibility for the proper operation of his product. If you live in such a country or state, you should not buy this product. If you nevertheless buy this program, Linguist Plus, Inc. considers that you are assuming complete responsibility concerning the effects of any possible inaccuracies of the program.

At the same time, if you do find that the programÕs operation shows a substantial difference from that described in the documentation, you are entitled to remedial action within 30 days. Proceed as follows:

1. Send a detailed description of the problem to Linguist Plus, Inc. Include pertinent signal(s) and describe your computer configuration (especially System used, sound input and output devices).
2. An attempt will be made to solve the problem within 30 days. If the problem cannot be solved, youÕre entitled to a full refund of the amount you paid for the program.

For a legally binding version of this word of caution, and for other legal aspects of the purchase and use of the Signalyze product, please refer to the Signalyze License Agreement. Upon request, you may obtain a free copy of the License Agreement prior to buying this product.

The Signalyze Demo

A demonstration version of Signalyze is available for test-driving. It is a fully functional version of the program, except that it will only open one file (included) and will not save signals.

The Signalyze Demo is available for download via the following link:

http://www.signalyze.com/downloads/Signalyze311demo.sea

More about NIH Image

Please note that Linguist Plus, Inc. has no links with ÒNIH ImageÓ or with its author, W. Rasband. We simply recommend the program as an freeware program to make 3D representations from Signalyze grayscale spectrograms.

To make a 3D spectrogram, you need a Macintosh capable of 256 colors or grayscales.

Proceed as follows:

1. Make a high-definition grayscale spectrogram with Signalyze.
2. Save the spectrogram in PICT format.
3. Load the spectrogram into NIH Image.
4. Frame the part of the spectrogram you want to convert into 3D.
5. Smooth the image.
6. Activate the Surface Plot menu command.

You can download the latest version of NIH Image via web browser at the NIH Image home page:

http://rsb.info.nih.gov/nih-image/

or via ftp at the Sunet NIH Image mirror site:

ftp://ftp.sunet.se/pub/mac/graphics/graphicsutil/nih-image

NIH Image is written by Wayne Rasband for the National Institutes of Health, Research Services Branch (NIMH).