Next: SL corpus design
Up: System design
Previous: Conclusion
In this section we will summarise the dimensions of a
requirement profile with keywords that will need to be regularly updated and illustrated by potential users.
A. Profile for speech recognition systems
Environment | Noise type (stationary,
speech-like?) |
| Signal-to-noise ratio |
| Utilisation conditions |
| Reverberation , acceleration, vibration |
Transducer | microphone (open, press-to-talk) |
| telephone (mobile?) |
Channel | bandwidth |
| distortion |
| echo, delay |
| |
Task | Lexicon size |
| Lexicon confusability |
| Perplexity factor |
| Dialogue size |
Speakers
| Speaker dependent/independent |
| number of training speakers available |
| number of training speakers to record |
| typology of users: |
| native/non-native ... |
| sex, |
| age range |
| physical/psychological state |
| social group |
| attitude (motivated...) |
| experienced/large public |
Speech | amplitude (quiet/normal/shout) |
| mode (isolated/continuous) |
| fluency (read , spontaneous ) |
| rate (slow, normal,fast) |
| conformity (speech/non-speech sounds) |
Vocabulary | Training material available |
| Training material to collect? |
| task specific/task independent |
| environment |
| transducer |
| channel |
| users |
| recency |
| control of vocabulary switching |
In-situ recording | possible/not |
(field data) | |
| |
Rejection mode | |
| |
Confusion matrix for the | |
application vocabulary | |
Language modelling | |
Channel and environment | |
adaptation | |
Task/Application | |
adaptation | |
Speech recognition and | |
application interfaces | |
Speech input and speech | |
signal acquisition | |
Cut-through versus | |
voice-stop (anticipation) | |
Error measure presentation | |
Error handling | |
Response time | |
Analogue connection | |
Analogue to digital | |
conversion | |
Digital interfaces | |
Noise conditions | |
PTT approval s | |
Controls: | |
Transducer level | |
settings (AGC) | |
| |
System performance | Recognition error rate |
| real-time aspects |
| system response-time |
| packaging aspects |
| (size, weight, power, cost) |
B. Profile for speaker verification systems
Error measure | |
Speaker verification error | |
Training | |
Exploitation | |
Text dependency / | |
independency | |
Speech quantity and quality | |
Adding/Removing speakers | |
C. Profile for a speaker identification system
Error measure | |
Speaker verification error | |
Training | |
Exploitation | |
Text dependency / | |
independency | |
Speech quantity and quality | |
Adding/Removing speakers | |
D. Profile for a speech synthesis system
Speech recording, storage, | |
and playback | |
Canned speech | |
Text-to-speech synthesis: | Linguistic part |
| Phonetic part |
| Acoustic module |
| Multi-linguality |
E. EAGLES SLWG keywords for speech synthesis profiles
A set of keywords, defined by the EAGLES Spoken Language Working Group 5
(meeting of November 1-2, 1993, Cambridge, chaired by Louis C.W. Pols)
related to system and application characterisations, is given below for
information.
1. | Text coverage | From concept, unlimited text, text |
| | interpretation (e.g. tables), |
| | (carrier phrases) plus keywords , |
| | punctuation , spell option, style specification |
| | language |
2. | Source | Coding and/or synthesis and/or canned speech
|
| | male/female/child |
| | style, emotion , rate, dialect |
| | adaptive to disturbances in channel or with user |
3. | Channel | High quality, telephone (handset, mobile, earphone), |
| | bandwidth , noise , reverberation , competing speech |
4. | User | Experience (one-time vs. multiple use), training, |
| | child-normal-elderly, (non-)native, 2nd language user, |
| | hearing impairment, (non-)cooperative |
5. | Application | Of reading-machine type OR of information-retrieval type? |
| | Field test of application itself (task completion) OR |
| | laboratory test of synthesis part alone, either with |
| | application-specific or application-independent tests. |
6. | Functional characteristics | Main emphasis on comprehension, intelligibility, naturalness , or otherwise? |
| | If intelligibility, then of all words or only of certain words? |
| | Consider separate evaluation of prosodic component |
| | If overall quality , then use set of scales (see above) |
| | Consider secondary tasks |
| | Performance in direct comparison, or in absolute sense, |
| | benchmarking |
| | How important are dialogue aspects (see other subgroups)? |
7. | Restrictions | Time, money, system availability |
8. | Alternatives and/or | |
| combined modes | Mouse, screen; |
| | add visual image; |
| | multimedia? |
| | Importance of hands-free, eyes-busy? |
9. | Technical details | Size, weight, price, interface, plug-in options, |
| | DSP board, modularity, diphone basis, |
| | options, hand-tuning, etc. |
F. System platform
Software aspects | Operating systems |
| Drivers |
| Application programming interfaces |
| Application generators |
| |
Hardware aspects | Platforms |
| Speech processing boards |
| Speech input/output interfaces |
| Connectivity |
| Real-time aspects |
| |
Planning for expansion | System simulation
and prototyping |
| Host-computer interfaces |
| Computer telephony integration |
| Multi-lingual aspects |
| System dimension configuration |
| Statistical tools |
| Blockage factor |
Ports | 10% | 5% | 2% | 1% | 0.5% | 0.1% |
4 | 2.05 | 1.52 | 1.09 | 0.87 | 0.70 | 0.44 |
5 | 2.88 | 2.22 | 1.66 | 1.36 | 1.13 | 0.76 |
6 | 3.76 | 2.96 | 2.28 | 1.91 | 1.62 | 1.15 |
7 | 4.67 | 3.74 | 2.94 | 2.50 | 2.16 | 1.58 |
8 | 5.60 | 4.54 | 3.63 | 3.13 | 2.73 | 2.05 |
9 | 6.55 | 5.37 | 4.34 | 3.78 | 3.33 | 2.56 |
10 | 7.51 | 6.22 | 5.08 | 4.46 | 3.96 | 3.09 |
11 | 8.49 | 7.08 | 5.84 | 5.16 | 4.61 | 3.65 |
12 | 9.47 | 7.95 | 6.62 | 5.88 | 5.28 | 4.23 |
13 | 10.47 | 8.83 | 7.41 | 6.61 | 5.96 | 4.83 |
14 | 11.47 | 9.73 | 8.20 | 7.35 | 6.66 | 5.45 |
15 | 12.48 | 10.63 | 9.01 | 8.11 | 7.38 | 6.08 |
16 | 13.50 | 11.54 | 9.83 | 8.87 | 8.10 | 6.72 |
17 | 14.52 | 12.46 | 10.66 | 9.65 | 8.83 | 7.38 |
18 | 15.55 | 13.38 | 11.49 | 10.44 | 9.58 | 8.05 |
19 | 16.58 | 14.31 | 12.33 | 11.23 | 10.33 | 8.72 |
20 | 17.61 | 15.25 | 13.18 | 12.03 | 11.09 | 9.41 |
21 | 18.65 | 16.19 | 14.04 | 12.84 | 11.86 | 10.11 |
22 | 19.69 | 17.13 | 14.90 | 13.65 | 12.64 | 10.81 |
23 | 20.74 | 18.08 | 15.76 | 14.47 | 13.42 | 11.52 |
24 | 21.78 | 19.03 | 16.63 | 15.29 | 14.20 | 12.24 |
25 | 22.83 | 19.99 | 17.50 | 16.12 | 15.00 | 12.97 |
26 | 23.88 | 20.94 | 18.38 | 16.96 | 15.80 | 13.70 |
27 | 24.94 | 21.90 | 19.26 | 17.80 | 16.60 | 14.44 |
28 | 26.00 | 22.87 | 20.15 | 18.64 | 17.41 | 15.18 |
29 | 27.05 | 23.83 | 21.04 | 19.49 | 18.22 | 15.93 |
30 | 28.11 | 24.80 | 21.93 | 20.34 | 19.04 | 16.68 |
31 | 29.17 | 25.77 | 22.83 | 21.19 | 19.86 | 17.44 |
32 | 30.23 | 26.75 | 23.73 | 22.05 | 20.68 | 18.20 |
33 | 31.30 | 27.72 | 24.63 | 22.91 | 21.51 | 18.97 |
34 | 32.36 | 28.70 | 25.53 | 23.77 | 22.34 | 19.74 |
35 | 33.43 | 29.68 | 26.43 | 24.64 | 23.17 | 20.52 |
36 | 34.50 | 30.66 | 27.34 | 25.51 | 24.01 | 21.30 |
37 | 35.57 | 31.64 | 28.25 | 26.38 | 24.85 | 22.08 |
38 | 36.64 | 32.63 | 29.17 | 27.25 | 25.69 | 22.86 |
39 | 37.71 | 33.61 | 30.08 | 28.13 | 26.54 | 23.65 |
40 | 38.79 | 34.60 | 31.00 | 29.01 | 27.38 | 24.44 |
41 | 39.86 | 35.59 | 31.92 | 28.89 | 28.23 | 25.24 |
42 | 40.94 | 36.58 | 32.84 | 30.77 | 29.08 | 26.04 |
43 | 42.01 | 37.57 | 33.76 | 31.66 | 29.94 | 26.84 |
44 | 43.09 | 38.56 | 34.68 | 32.54 | 30.80 | 27.64 |
45 | 44.16 | 39.55 | 35.61 | 33.43 | 31.66 | 28.45 |
46 | 45.24 | 40.54 | 36.53 | 34.32 | 32.52 | 29.26 |
47 | 46.32 | 41.54 | 37.46 | 35.21 | 33.38 | 30.07 |
48 | 47.40 | 42.54 | 38.39 | 36.11 | 34.25 | 30.88 |
49 | 48.48 | 43.54 | 39.32 | 37.00 | 35.11 | 31.69 |
50 | 49.56 | 44.53 | 40.25 | 37.90 | 35.98 | 32.51 |
51 | 50.60 | 45.50 | 41.20 | 38.80 | 36.85 | 33.30 |
52 | 51.70 | 46.50 | 42.10 | 39.70 | 37.72 | 34.20 |
53 | 52.80 | 47.50 | 43.10 | 40.60 | 38.60 | 35.00 |
54 | 53.90 | 48.50 | 44.00 | 41.50 | 39.47 | 35.80 |
55 | 55.00 | 49.50 | 44.90 | 42.40 | 40.35 | 36.60 |
56 | 56.10 | 50.50 | 45.90 | 43.30 | 41.23 | 37.50 |
57 | 57.10 | 51.50 | 46.80 | 44.20 | 42.11 | 38.30 |
58 | 58.20 | 52.60 | 47.80 | 45.10 | 42.99 | 39.10 |
59 | 59.30 | 53.60 | 48.70 | 46.00 | 43.88 | 40.00 |
60 | 60.40 | 54.60 | 49.60 | 46.90 | 44.76 | 40.80 |
61 | 61.50 | 55.60 | 50.60 | 47.90 | 45.64 | 41.60 |
62 | 62.60 | 56.60 | 51.50 | 48.80 | 46.53 | 42.50 |
63 | 63.70 | 57.60 | 52.50 | 49.70 | 47.42 | 43.30 |
64 | 64.80 | 58.60 | 53.40 | 50.60 | 48.31 | 44.20 |
65 | 65.80 | 59.60 | 54.40 | 51.50 | 49.19 | 45.00 |
66 | 66.90 | 60.60 | 55.30 | 52.40 | 50.09 | 45.80 |
67 | 68.00 | 61.60 | 56.30 | 53.40 | 50.98 | 46.70 |
68 | 69.10 | 62.60 | 57.20 | 54.30 | 51.87 | 47.50 |
69 | 70.20 | 63.70 | 58.20 | 55.20 | 52.77 | 48.40 |
70 | 71.30 | 64.70 | 59.10 | 56.10 | 53.66 | 49.20 |
71 | 72.40 | 65.70 | 60.10 | 57.00 | 54.56 | 50.10 |
72 | 73.50 | 66.70 | 61.00 | 58.00 | 55.46 | 50.90 |
73 | 74.60 | 67.70 | 62.00 | 58.90 | 56.35 | 51.80 |
74 | 75.60 | 68.70 | 62.90 | 59.80 | 57.25 | 52.70 |
75 | 76.70 | 69.70 | 63.90 | 60.70 | 58.15 | 53.50 |
76 | 77.80 | 70.80 | 64.90 | 61.70 | 59.05 | 54.40 |
77 | 78.90 | 71.80 | 65.80 | 62.60 | 59.96 | 55.20 |
78 | 80.00 | 72.80 | 66.80 | 63.50 | 60.86 | 56.10 |
79 | 81.10 | 73.80 | 67.70 | 64.40 | 61.77 | 57.00 |
80 | 82.20 | 74.80 | 68.70 | 65.40 | 62.67 | 57.80 |
81 | 83.30 | 75.80 | 69.60 | 66.30 | 63.57 | 58.70 |
82 | 84.40 | 76.90 | 70.60 | 67.20 | 64.48 | 59.50 |
83 | 85.50 | 77.90 | 71.60 | 68.20 | 65.39 | 60.40 |
84 | 86.60 | 78.90 | 72.50 | 69.10 | 66.29 | 61.30 |
85 | 87.70 | 79.90 | 73.50 | 70.00 | 67.20 | 62.10 |
86 | 88.80 | 80.90 | 74.50 | 70.90 | 68.11 | 63.00 |
87 | 89.90 | 82.00 | 75.40 | 71.90 | 69.02 | 63.90 |
88 | 91.00 | 83.00 | 76.40 | 72.80 | 69.93 | 64.70 |
89 | 92.10 | 84.00 | 77.30 | 73.70 | 70.85 | 65.60 |
90 | 93.10 | 85.00 | 78.30 | 74.70 | 71.75 | 66.50 |
91 | 94.20 | 86.00 | 79.30 | 75.60 | 72.67 | 67.40 |
92 | 95.30 | 87.10 | 80.20 | 76.60 | 73.58 | 68.20 |
93 | 96.40 | 88.10 | 81.20 | 77.50 | 74.49 | 69.10 |
94 | 97.50 | 89.10 | 82.20 | 78.40 | 75.41 | 70.00 |
95 | 98.60 | 90.10 | 83.10 | 79.40 | 76.33 | 70.90 |
96 | 99.70 | 91.10 | 84.10 | 80.30 | 77.24 | 71.70 |
97 | 100.80 | 92.20 | 85.10 | 81.20 | 78.16 | 72.60 |
98 | 101.90 | 93.20 | 86.00 | 82.20 | 79.07 | 73.50 |
99 | 103.00 | 94.20 | 87.00 | 83.10 | 79.99 | 74.40 |
100 | 104.10 | 95.20 | 88.00 | 84.10 | 80.91 | 75.20 |
101 | 105.20 | 96.30 | 88.90 | 85.00 | 81.83 | 76.10 |
102 | 106.30 | 97.30 | 89.90 | 85.90 | 82.75 | 77.00 |
103 | 107.40 | 98.30 | 90.90 | 86.90 | 83.67 | 77.90 |
104 | 108.50 | 99.30 | 91.90 | 87.80 | 84.59 | 78.80 |
105 | 109.60 | 100.40 | 92.80 | 88.80 | 85.51 | 79.70 |
106 | 110.70 | 101.40 | 93.80 | 89.70 | 86.43 | 80.50 |
107 | 111.80 | 102.40 | 94.80 | 90.70 | 87.36 | 81.40 |
108 | 112.90 | 103.40 | 95.70 | 91.60 | 88.28 | 82.30 |
109 | 114.00 | 104.50 | 96.70 | 92.50 | 89.20 | 83.20 |
110 | 115.10 | 105.50 | 97.70 | 93.50 | 90.12 | 84.10 |
111 | 116.20 | 106.50 | 98.70 | 94.40 | 91.05 | 85.00 |
112 | 117.30 | 107.50 | 99.60 | 95.40 | 91.97 | 85.90 |
113 | 118.40 | 108.60 | 100.60 | 96.30 | 92.89 | 86.70 |
114 | 119.50 | 109.60 | 101.60 | 97.30 | 93.82 | 87.60 |
115 | 120.60 | 110.60 | 102.50 | 98.20 | 94.74 | 88.50 |
116 | 121.70 | 111.70 | 103.50 | 99.20 | 95.67 | 89.40 |
117 | 122.80 | 112.70 | 104.50 | 100.10 | 96.60 | 90.30 |
118 | 123.90 | 113.70 | 105.50 | 101.10 | 97.52 | 91.20 |
119 | 125.00 | 114.70 | 106.40 | 102.00 | 98.45 | 92.10 |
120 | 126.10 | 115.80 | 107.40 | 103.00 | 99.38 | 93.00 |
Table 2.5: Erlang B carried-traffic table
Next: SL corpus design
Up: System design
Previous: Conclusion
EAGLES SWLG SoftEdition, May 1997. Get the book...