Whisper gives different results based on the platform

[Tharuja]

I have tried the Whisper model with numpy==1.24.3 version in both Linux-AMD64 and MacOs M1-ARM systems.. Both give two different results as shown below (with different no_speech_prob values)... What is the reason for this? I used Python 3.8 version.

Results in MacOs M1-ARM

{ "text": " The number one biggest difference between the bacteria in your body and the cells making of your body are these tiny cellular components called organelles. You've actually learned a lot about organelles in other lessons without knowing it." "language": "en", "segments": [ { "id": 0, "seek": 0, "start": 0.0, "end": 5.0, "text": " The number one biggest difference between the bacteria in your body and the cells making", "tokens": [440, 1230, 472, 3880, 2649, 1296, 264, 11763, 294, 428, 1772, 293, 264, 5438, 1455] "temperature": 0.0, "avg_logprob": -0.23756315901472763, "compression_ratio": 1.5868544600938967, "no_speech_prob": 0.018022961914539337, }, { "id": 1, "seek": 0, "start": 5.0, "end": 9.24, "text": " of your body are these tiny cellular components called organelles.", "tokens": [295, 428, 1772, 366, 613, 5870, 29267, 6677, 1219, 1798, 19787,13] "temperature": 0.0, "avg_logprob": -0.23756315901472763, "compression_ratio": 1.5868544600938967, "no_speech_prob": 0.018022961914539337, }, { "id": 2, "seek": 0, "start": 9.24, "end": 14.08, "text": " You've actually learned a lot about organelles in other lessons without knowing it." "tokens": [509, 600, 767, 3264, 257, 688, 466,1798, 19787, 294, 661, 8820, 1553, 5276, 309, 13] "temperature": 0.0, "avg_logprob": -0.23756315901472763, "compression_ratio": 1.5868544600938967, "no_speech_prob": 0.018022961914539337, }, ], }

Results in Linux-AMD64 :-

{ 'text': " The number one biggest difference between the bacteria in your body and the cells making of your body are these tiny cellular components called organelles. You've actually learned a lot about organelles in other lessons without knowing it.", 'language': 'en' 'segments': [ { 'id': 0, 'seek': 0, 'start': 0.0, 'end': 5.0, 'text': ' The number one biggest difference between the bacteria in your body and the cells making', 'tokens': [50364, 440, 1230, 472, 3880, 2649, 1296, 264, 11763, 294, 428, 1772, 293, 264, 5438, 1455, 50614], 'temperature': 0.0, 'avg_logprob': -0.25806166330973307, 'compression_ratio': 1.610878661087866, 'no_speech_prob': 0.002833886770531535 }, { 'id': 1, 'seek': 0, 'start': 5.0, 'end': 9.24, 'text': ' of your body are these tiny cellular components called organelles.', 'tokens': [50614, 295, 428, 1772, 366, 613, 5870, 29267, 6677, 1219, 1798, 19787, 13, 50826], 'temperature': 0.0, 'avg_logprob': -0.25806166330973307, 'compression_ratio': 1.610878661087866, 'no_speech_prob': 0.002833886770531535 }, { 'id': 2, 'seek': 0, 'start': 9.24, 'end': 14.08, 'text': " You've actually learned a lot about organelles in other lessons without knowing it.", 'tokens': [50826, 509, 600, 767, 3264, 257, 688, 466, 1798, 19787, 294, 661, 8820, 1553, 5276, 309, 13, 51068], 'temperature': 0.0, 'avg_logprob': -0.25806166330973307, 'compression_ratio': 1.610878661087866, 'no_speech_prob': 0.002833886770531535 }, ], }

[phineas-pta]

it's the default behaviour see #81