You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
python3 -m whisper_bidec "what's the temperature of the EcoBee.wav"
55
+
```
56
+
57
+
This outputs CSV with the format `wav file|text without bias|text with bias` like:
58
+
59
+
```csv
60
+
what's the temperature of the EcoBee.wav|What's the temperature of the incubi?|What's the temperature of the incubi?
61
+
```
62
+
63
+
Without bias, the WAV file is incorrectly transcribed as "What's the temperature of the **incubi**?"
64
+
65
+
Let's add a few example sentences that will bias Whisper towards the "EcoBee" device:
66
+
67
+
```sh
68
+
cat > example_sentences.txt <<EOF
69
+
What's the temperature of the EcoBee?
70
+
What is the temperature of the EcoBee?
71
+
EOF
72
+
```
73
+
74
+
Now we can see the corrected transcript:
75
+
76
+
```sh
77
+
python3 -m whisper_bidec --text example_sentences.txt "what's the temperature of the EcoBee.wav"
78
+
what's the temperature of the EcoBee.wav|What's the temperature of the incubi?|What's the temperature of the EcoBee?
79
+
```
80
+
81
+
The bias can be adjusted with `--bias-towards-lm <BIAS>` which defaults to 0.5. Increasing this value will bias Whisper more towards the example sentences.
0 commit comments