Table 2.
Comparison of the Features of the Most Popular Continuous Speech Recognition Systems
| DragonSystems' Naturally Speaking | IBM ViaVoice Gold | Phillips SpeechMagic | |
|---|---|---|---|
| “Active” vocabulary | 30,000-55,000 words | 22,000-64,000 words | 64,000 words |
| Speed | 100-160+ wpm | 125 wpm | Batch mode |
| Synchronous correction | Yes | Yes | Yes |
| Correction by voice only | Yes | No | Yes |
| Document navigation | Yes | No | Yes* |
| On-the-spot vocabulary builder | Yes | Yes | No |
| Training time | 20-60 minutes | A few minutes | 10+ minutes |
| Peak accuracy (per vendor) | 98% | 97% | 100% |
| Speaker independent | No | Yes† | No |
| Speaker adaptive | Yes | Yes | Yes |
| Learns words from input documents | Yes | Yes | No |
| Facility for “macros” | Yes | Yes | No |
| Dictate directly into other applications | Yes | Yes | Yes‡ |
| Software developers kit available | Yes | Yes | Yes |
| Hardware requirements (Windows 95) | Pentium 133, 32-MB RAM | Pentium 150 MMX, 32-MB RAM | Pentium 166, 64-MB RAM |
| Software price | $100-$695 | $100 | $6000 |
*
The SpeechMagic system is a batch-mode-only system. Thus, documents cannot be navigated at dictation time because they have not yet been created.
†
IBM supplies a “user wizard” that prompts the user to select from preset voice profiles. This is not true speaker independence, and such functionality also exists for the DragonSystems package. The user has to spend about five minutes with this wizard at the time of initial enrollment. This may have changed with ViaVoice 98.
‡
Using the software toolkit the user can create hooks to send dictations anywhere. This is also true for the other two applications.