Because our Streaming Speech publications are so innovative, many people have asked questions about the underlying concepts, and in particular about the generation and use of transcripts. Below are some questions and answers which are of general interest.

1. What do you mean by 'Discourse Features'?
2. How accurate are the transcriptions?
3. How can Streaming Speech be used by a 'tone deaf' teacher?
4. I disagree with your transcription!
5. What are speech units?
6. Speed: why do you measure speed in words per minute?
7. Speed: why are the speeds so precise?

In Streaming Speech I use the term ‘Discourse Features’ to refer to those characteristics which speech has when it is used in communication. In normal communication, words do not usually occur alone, they occur with other words in speech units: words are run together in ways which pull them out of shape – that is, they become unlike their dictionary forms. In speech units, the sound-shapes of words are affected by a number of factors: the position of prominences, the occurrence of tone, and key; and the rhythms, speed, and volume of the current and surrounding speech units. ‘Discourse Features’ are those characteristics of the stream of speech that dissolve and re-shape the dictionary pronunciation of words. Discourse Features are dealt with in every Section 3 of Chapters 1-8, and in Chapter 10.

They are as accurate as I can make them. At the time of writing I have nearly 20 years experience of doing transcriptions using the Window on Speech (Chapter 10). I have tested many of the recordings on colleagues, and students, and I am confident that the majority of the transcriptions are accurate.

However, it is important to remember that we are dealing with fast spontaneous speech, which is (within limits) infinitely variable in its streamlike qualities: it is subject to microvariations in volume, pitch, etc. It is therefore difficult, if not impossible, to be certain about every step in the analysis into speech units. When dealing with spontaneous speech, it is necessary to be able to tolerate uncertainty.

Many people tell me that they are ‘tone deaf’: they say that they cannot hear prominences, non-prominences, falling or rising tones, that they cannot hear steps up or steps down in the voice. When they tell me they are tone deaf, I don't believe them. Among the hundreds of students that I have worked with over the years I have only met one who is close to being tone deaf. All the others were able to learn to hear prominences, tones, and key.

We both hear the same speech unit, the same stretch of the stream of speech – but we are categorising the moments of the speech unit differently. They are focusing on a part of the speech unit which undeniably exists (most falls are preceded by a slight jump upwards) but is is not the part which is important in the descriptive system (Discourse Intonation) we are using. For the descriptive system, the important part is the fall – this is the speaker’s choice – and the small jump upwards that immediately precedes the fall is not important in the descriptive system. As an experienced transcriber using this system I make myself less sensitive to these small jumps upwards, and more sensitive to what happens after the jump – the falling tone.

I have had ear-training, and I have used the descriptive system of speech units for nearly twenty years. I have been trained to hear speech in a certain way. Such training requires time, and some effort. All such training, whatever system you use, requires you to become more sensitive to those parts of the stream of speech that the descriptive system recognises, and it trains you to de-sensitise youself to those parts of the stream of speech that it does not recognise.

I do not believe that anyone is tone deaf – but I admit that many people believe so strongly that they are tone deaf that they find it very difficult to learn to hear the Discourse Features of speech. Most of the time, belief is the problem, not tone deafness.

There are a number of reasons why you may disagree with my transcriptions.

• We are listening to the recordings on different equipment. I made the transcriptions using CD-quality recordings and playback: perhaps your equipment is not as good as mine, or perhaps it is superior to mine: in either case it is likely that it is different equipment. Different equipment will produce different qualities of sound.
   
• In the process of making Streaming Speech, these recordings were compressed into MP3 files as part of Macromedia Flash movies. There is therefore less information in the sound recordings that you hear, than in the recordings on which I did my analysis into speech units.
   
• You may not have developed your ear to hear the categories of the descriptive system that I use. This system, which is based on David Brazil's Discourse Intonation, includes five sizes of speech-unit, five tones, prominence & non-prominence, and high and low key (The Window on Speech, cf. Chapter 10). If this is the case, be patient: in cases of disagreement, think to yourself 'OK, I disagree here, but this is how Streaming Speech analyses it'.
   
• You may be wrong, or you may have a legitimate, other, opinion. It is perfectly possible for experts, even experts using the same descriptive system, to disagree.
   
• I may be wrong. It is quite likely that I have made a few mistakes.

Speech units are stretches of speech, usually larger than a word, which have their own rhythms, tones, and other features which make speech stream-like. Speech units are very important in Streaming Speech: they are the units that we work with in teaching Listening and Pronunciation. Speech units are dealt with in every Section 3 of Chapters 1-8 and in Chapter 10.

There are two main ways of measuring speed of speech: words per minute, and syllables per second. Both ways of measuring speed have their advantages and disadvantages. In language teaching, it is common to deal in words per minute (wpm). In research, it is more common to deal in syllables per second (sps). In Streaming Speech I use words per minute. This works well most of the time, but in Chapter 7, where Rachel (British/Irish version) and Travis (American/Canadian version) use words with many syllables, I thought it worth showing the differences that would result from computing speed using two methods, words per minute, and syllables per minute. See Chapter 7, Section 2.

The speed of each speech unit is given as a three-figure digit after each speech unit. For example, the nineteenth speech unit in Corony’s recording has the three digits 282 following it, indicating that she speaks these words at a speed of 282 words per minute.

All figures given for speed of speech were automatically generated by software known as ‘Motormouth’ developed by myself and Trevor Batchelor of the University of Birmingham. The software computes the figures for speed by following four steps:

• it takes the length of each speech unit (in this case 1.28 seconds)
   
• it counts the number of words (in this case, 6 words)
   
• it divides the length by the number of words in order to give the average duration for each word in the speech unit (in this case, 1.28/6 = 0.213seconds)
   
• it divides 60 seconds by this result (60/0.213 = 281.6) giving a speed in words per minute

The figures given for speed in Streaming Speech are over-precise. I suggest therefore, that when you look at the numbers for speed, that you discount the last digit. So that when looking at Corony’s 019, you regard it as being spoken at 280 words per minute, not at 282 words per minute. This practice, of rounding down to the nearest '0' was adopted for the American/Canadian version