{"text": "選擇日語的專用模型,選擇下日語,勾選上AI智能拆函。等待2分鐘,2分鐘的日語視頻就轉錄好了。浩叔開發的字幕軟件Hawang發布威死版本了。這個版本最主要的更新就是增加了日語轉錄與翻譯的功能。那麼轉錄的質量如何呢?播放一下看一下。可以看到,準確率還是非常不錯的,猜寒也比較合理。人聲與音頻的對齊率非常高。以方劇轉錄的場景為例,軟件重點解決了如下三個遇到的核心問題。如何提高日語轉錄的準確率與穩定性?Hawang軟件,契怨的Vespa的模型。使怨的Parageter這個日語專用的模型。這是經過浩叔認真評測的選擇。詳細的可以看我上一期的視頻。睿頻2026年日語ASR開源模型。經過評測,Parageter模型無論在識別的準確率、穩定性、詞語級的對齊都好於Vespa。如何提高轉錄的速度呢?日語方距都是長視頻。Hawang轉錄非常的快。30分鐘的視頻,3分鐘就可以完成轉錄了。一般性能的顯卡也能實現高速的轉錄。Hawang軟件針對長音頻音化。2小時的日據轉錄之後,也不會出現局部的崩壞、時間搓飄移等問題。Hawang還支持批量的轉錄。你可以一次性轉錄一批方劇,不願等待時間一個個的操作。第三個問題就是,如何實現快速的糾錯?日語的AI識別是必然有錯誤的。Hawang作為專業的字幕軟件,在字幕糾錯上面效率非常的高。帶有AI識別與糾錯的功能,你可以利用AI快速找出錯誤點並修正,並且支持將糾錯詞一鍵配置到樂詞表中。帶有AI樂詞替換的功能,你可以把人名與專有名詞配置在樂詞表中。實現轉錄時的自動替換。帶有日語文稿匹配的功能。比如我會使用JomLite轉錄出文字稿。JomLite的轉錄很卷,但時間戳不行。將文稿捻貼到文稿匹配中,就可以實現快速的糾錯。接下來,來完整演示一下軟件的操作。第一次試驗,請先下載日語專業模型。然後將文件導入到軟件中。點擊本地轉錄。選擇日語專業模型。選擇日語。勾選上AI智能拆行。可以解決轉錄的日語容易出長距的問題。調整合適的單行字幕最大長度。點擊執行轉錄。等待一會兒日語字幕就轉錄完成了。下一步你可以點擊AI實錯與糾錯。這是日語的實錯提示詞。你可以添加自己的實錯的規則。執行完成之後,會出現二次確認的表。檢查後沒有問題。你就可以將常見的錯誤。一鍵批量添加到熱池表中。點擊確認後,程序會自動完成批量的替換。你可以到設置的頁面中,找到熱池表。切換到日語。你就會看到熱池了。在這裡你可以管理你的熱池。然後在轉錄的時候,開啟AI熱池替換開關就可以了。最後,有雙語字幕的需求的。可以點擊下雙語字幕。上行可以選擇日語,下行選擇中文。點擊開始處理。等待一會兒,雙語的字幕就生成好了。點擊導出,就可以將ASR的字幕導出了。我是浩叔,我們下一個視頻見。", "segments": [{"id": 0, "seek": 0, "start": 0.0, "end": 9.6, "text": "選擇日語的專用模型,選擇下日語,勾選上AI智能拆函。等待2分鐘,2分鐘的日語視頻就轉錄好了。", "tokens": [50365, 21625, 13167, 229, 6890, 31348, 1546, 46072, 9254, 41908, 39823, 11, 21625, 13167, 229, 4438, 6890, 31348, 11, 7978, 122, 21625, 5708, 48698, 5094, 118, 8225, 6852, 228, 6336, 121, 1543, 10187, 18390, 17, 32611, 11, 17, 32611, 1546, 6890, 31348, 27333, 29940, 3111, 32739, 9567, 226, 12621, 1543, 50845], "temperature": 0, "avg_logprob": -0.22551321782985656, "compression_ratio": 1.231372549019608, "no_speech_prob": 9.123923838671999e-12}, {"id": 1, "seek": 0, "start": 11.4, "end": 24.0, "text": "浩叔開發的字幕軟件Hawang發布威死版本了。這個版本最主要的更新就是增加了日語轉錄與翻譯的功能。那麼轉錄的質量如何呢?播放一下看一下。", "tokens": [50935, 11038, 102, 39936, 8949, 14637, 1546, 22381, 40211, 14454, 253, 20485, 39, 1607, 656, 14637, 34688, 19976, 223, 16989, 42096, 8802, 2289, 1543, 6287, 42096, 8802, 8661, 13557, 4275, 1546, 19002, 12560, 5620, 24228, 252, 9990, 2289, 6890, 31348, 32739, 9567, 226, 34127, 42716, 11823, 107, 1546, 32311, 8225, 1543, 20504, 32739, 9567, 226, 1546, 43450, 26748, 43526, 6240, 30, 49993, 12744, 8861, 28324, 1543, 51565], "temperature": 0, "avg_logprob": -0.22551321782985656, "compression_ratio": 1.231372549019608, "no_speech_prob": 9.123923838671999e-12}, {"id": 2, "seek": 3000, "start": 30.0, "end": 58.0, "text": "可以看到,準確率還是非常不錯的,猜寒也比較合理。人聲與音頻的對齊率非常高。以方劇轉錄的場景為例,軟件重點解決了如下三個遇到的核心問題。", "tokens": [50365, 6723, 18032, 11, 28220, 24293, 44866, 25583, 14392, 30967, 1546, 11, 25266, 250, 4510, 240, 6404, 25174, 14245, 13876, 1543, 4035, 29454, 34127, 18034, 29940, 1546, 2855, 44628, 232, 44866, 14392, 12979, 1543, 3588, 9249, 5087, 229, 32739, 9567, 226, 1546, 21446, 50218, 6344, 17797, 11, 14454, 253, 20485, 12624, 8216, 17278, 33540, 2289, 8238, 4438, 10960, 3338, 3330, 229, 4511, 1546, 8987, 116, 7945, 17197, 1543, 51765], "temperature": 0, "avg_logprob": -0.2464776912205656, "compression_ratio": 1.1142857142857143, "no_speech_prob": 1.7101292604237983e-11}, {"id": 3, "seek": 5800, "start": 58.0, "end": 64.0, "text": "如何提高日語轉錄的準確率與穩定性?", "tokens": [50365, 43526, 20949, 12979, 6890, 31348, 32739, 9567, 226, 1546, 28220, 24293, 44866, 34127, 13740, 102, 12088, 21686, 30, 50665], "temperature": 0, "avg_logprob": -0.2600473553946849, "compression_ratio": 1.0945273631840795, "no_speech_prob": 1.8057935355364485e-11}, {"id": 4, "seek": 5800, "start": 64.0, "end": 76.0, "text": "Hawang軟件,契怨的Vespa的模型。使怨的Parageter這個日語專用的模型。這是經過浩叔認真評測的選擇。詳細的可以看我上一期的視頻。", "tokens": [50665, 39, 1607, 656, 14454, 253, 20485, 11, 1881, 239, 3757, 101, 1546, 53, 279, 4306, 1546, 41908, 39823, 1543, 22982, 3757, 101, 1546, 46087, 559, 2398, 6287, 6890, 31348, 46072, 9254, 1546, 41908, 39823, 1543, 16600, 15145, 8816, 11038, 102, 39936, 22041, 6303, 39115, 9592, 105, 1546, 21625, 13167, 229, 1543, 6716, 111, 41588, 1546, 6723, 4200, 1654, 5708, 2257, 16786, 1546, 27333, 29940, 1543, 51265], "temperature": 0, "avg_logprob": -0.2600473553946849, "compression_ratio": 1.0945273631840795, "no_speech_prob": 1.8057935355364485e-11}, {"id": 5, "seek": 7600, "start": 76.0, "end": 92.0, "text": "睿頻2026年日語ASR開源模型。經過評測,Parageter模型無論在識別的準確率、穩定性、詞語級的對齊都好於Vespa。如何提高轉錄的速度呢?日語方距都是長視頻。", "tokens": [50365, 14221, 123, 29940, 2009, 10880, 5157, 6890, 31348, 3160, 49, 8949, 47402, 41908, 39823, 1543, 15145, 8816, 39115, 9592, 105, 11, 46087, 559, 2398, 41908, 39823, 16976, 29340, 3581, 43143, 16158, 1546, 28220, 24293, 44866, 1231, 13740, 102, 12088, 21686, 1231, 6716, 252, 31348, 35315, 1546, 2855, 44628, 232, 7182, 2131, 19488, 53, 279, 4306, 1543, 43526, 20949, 12979, 32739, 9567, 226, 1546, 31217, 13127, 6240, 30, 6890, 31348, 9249, 6563, 251, 22796, 15353, 27333, 29940, 1543, 51165], "temperature": 0, "avg_logprob": -0.1833317839069131, "compression_ratio": 1.0357142857142858, "no_speech_prob": 1.907278675272739e-11}, {"id": 6, "seek": 9200, "start": 92.0, "end": 102.0, "text": "Hawang轉錄非常的快。30分鐘的視頻,3分鐘就可以完成轉錄了。一般性能的顯卡也能實現高速的轉錄。", "tokens": [50365, 39, 1607, 656, 32739, 9567, 226, 48263, 10251, 1543, 3446, 32611, 1546, 27333, 29940, 11, 18, 32611, 36781, 41509, 32739, 9567, 226, 2289, 1543, 2257, 49640, 21686, 8225, 1546, 46605, 32681, 6404, 8225, 10376, 9581, 12979, 31217, 1546, 32739, 9567, 226, 1543, 50865], "temperature": 0, "avg_logprob": -0.13298701533564813, "compression_ratio": 1.3046594982078854, "no_speech_prob": 2.277751771917469e-11}, {"id": 7, "seek": 9200, "start": 102.0, "end": 112.0, "text": "Hawang軟件針對長音頻音化。2小時的日據轉錄之後,也不會出現局部的崩壞、時間搓飄移等問題。", "tokens": [50865, 39, 1607, 656, 14454, 253, 20485, 5873, 251, 2855, 15353, 18034, 29940, 18034, 23756, 1543, 17, 46997, 1546, 6890, 36841, 32739, 9567, 226, 26362, 11, 6404, 21121, 7781, 9581, 34703, 13470, 1546, 49378, 102, 12829, 252, 1231, 20788, 22220, 241, 11808, 226, 8204, 119, 10187, 17197, 1543, 51365], "temperature": 0, "avg_logprob": -0.13298701533564813, "compression_ratio": 1.3046594982078854, "no_speech_prob": 2.277751771917469e-11}, {"id": 8, "seek": 9200, "start": 112.0, "end": 119.0, "text": "Hawang還支持批量的轉錄。你可以一次性轉錄一批方劇,不願等待時間一個個的操作。", "tokens": [51365, 39, 1607, 656, 7824, 35488, 3416, 117, 26748, 1546, 32739, 9567, 226, 1543, 42766, 27505, 21686, 32739, 9567, 226, 2257, 3416, 117, 9249, 5087, 229, 11, 1960, 20692, 10187, 18390, 20788, 8990, 3338, 1546, 13167, 235, 11914, 1543, 51715], "temperature": 0, "avg_logprob": -0.13298701533564813, "compression_ratio": 1.3046594982078854, "no_speech_prob": 2.277751771917469e-11}, {"id": 9, "seek": 11900, "start": 119.0, "end": 131.0, "text": "第三個問題就是,如何實現快速的糾錯?日語的AI識別是必然有錯誤的。Hawang作為專業的字幕軟件,在字幕糾錯上面效率非常的高。", "tokens": [50365, 35878, 3338, 17197, 5620, 11, 43526, 10376, 9581, 10251, 31217, 1546, 15239, 122, 13133, 30, 6890, 31348, 1546, 48698, 43143, 16158, 1541, 28531, 5823, 2412, 13133, 3549, 97, 1546, 1543, 39, 1607, 656, 11914, 6344, 46072, 27119, 1546, 22381, 40211, 14454, 253, 20485, 11, 3581, 22381, 40211, 15239, 122, 13133, 49750, 43076, 44866, 48263, 12979, 1543, 50965], "temperature": 0, "avg_logprob": -0.08038865451155038, "compression_ratio": 1.3794326241134751, "no_speech_prob": 2.0834981309669764e-11}, {"id": 10, "seek": 11900, "start": 131.0, "end": 148.0, "text": "帶有AI識別與糾錯的功能,你可以利用AI快速找出錯誤點並修正,並且支持將糾錯詞一鍵配置到樂詞表中。帶有AI樂詞替換的功能,你可以把人名與專有名詞配置在樂詞表中。", "tokens": [50965, 32379, 2412, 48698, 43143, 16158, 34127, 15239, 122, 13133, 1546, 32311, 8225, 11, 42766, 23700, 9254, 48698, 10251, 31217, 25085, 7781, 13133, 3549, 97, 8216, 35934, 7792, 106, 15789, 11, 35934, 20334, 35488, 34168, 15239, 122, 13133, 6716, 252, 2257, 41668, 113, 38846, 34719, 4511, 34043, 6716, 252, 17571, 5975, 1543, 32379, 2412, 48698, 34043, 6716, 252, 9531, 123, 36338, 1546, 32311, 8225, 11, 42766, 16075, 4035, 15940, 34127, 46072, 2412, 15940, 6716, 252, 38846, 34719, 3581, 34043, 6716, 252, 17571, 5975, 1543, 51815], "temperature": 0, "avg_logprob": -0.08038865451155038, "compression_ratio": 1.3794326241134751, "no_speech_prob": 2.0834981309669764e-11}, {"id": 11, "seek": 14800, "start": 148.0, "end": 160.0, "text": "實現轉錄時的自動替換。帶有日語文稿匹配的功能。比如我會使用JomLite轉錄出文字稿。JomLite的轉錄很卷,但時間戳不行。", "tokens": [50365, 10376, 9581, 32739, 9567, 226, 6611, 1546, 9722, 12572, 9531, 123, 36338, 1543, 32379, 2412, 6890, 31348, 17174, 10415, 123, 9937, 117, 38846, 1546, 32311, 8225, 1543, 36757, 1654, 6236, 22982, 9254, 41, 298, 43, 642, 32739, 9567, 226, 7781, 17174, 22381, 10415, 123, 1543, 41, 298, 43, 642, 1546, 32739, 9567, 226, 4563, 5322, 115, 11, 8395, 20788, 1486, 111, 24151, 1543, 50965], "temperature": 0, "avg_logprob": -0.19620394706726074, "compression_ratio": 1.1709844559585492, "no_speech_prob": 2.228649730040555e-11}, {"id": 12, "seek": 14800, "start": 160.0, "end": 164.0, "text": "將文稿捻貼到文稿匹配中,就可以實現快速的糾錯。", "tokens": [50965, 34168, 17174, 10415, 123, 26075, 119, 11561, 120, 4511, 17174, 10415, 123, 9937, 117, 38846, 5975, 11, 36781, 10376, 9581, 10251, 31217, 1546, 15239, 122, 13133, 1543, 51165], "temperature": 0, "avg_logprob": -0.19620394706726074, "compression_ratio": 1.1709844559585492, "no_speech_prob": 2.228649730040555e-11}, {"id": 13, "seek": 16400, "start": 164.0, "end": 180.0, "text": "接下來,來完整演示一下軟件的操作。第一次試驗,請先下載日語專業模型。然後將文件導入到軟件中。點擊本地轉錄。選擇日語專業模型。選擇日語。勾選上AI智能拆行。", "tokens": [50365, 40012, 11, 3763, 14128, 27662, 31382, 25696, 8861, 14454, 253, 20485, 1546, 13167, 235, 11914, 1543, 18049, 9487, 22099, 24023, 245, 11, 16302, 10108, 4438, 12290, 231, 6890, 31348, 46072, 27119, 41908, 39823, 1543, 10213, 34168, 17174, 20485, 25647, 14028, 4511, 14454, 253, 20485, 5975, 1543, 8216, 37813, 8802, 10928, 32739, 9567, 226, 1543, 21625, 13167, 229, 6890, 31348, 46072, 27119, 41908, 39823, 1543, 21625, 13167, 229, 6890, 31348, 1543, 7978, 122, 21625, 5708, 48698, 5094, 118, 8225, 6852, 228, 8082, 1543, 51165], "temperature": 0, "avg_logprob": -0.09831014899320381, "compression_ratio": 1.2598870056497176, "no_speech_prob": 2.134322059366145e-11}, {"id": 14, "seek": 18000, "start": 180.0, "end": 192.0, "text": "可以解決轉錄的日語容易出長距的問題。調整合適的單行字幕最大長度。點擊執行轉錄。等待一會兒日語字幕就轉錄完成了。", "tokens": [50365, 6723, 17278, 33540, 32739, 9567, 226, 1546, 6890, 31348, 49212, 7781, 15353, 6563, 251, 1546, 17197, 1543, 32943, 27662, 14245, 3330, 102, 1546, 24227, 8082, 22381, 40211, 8661, 3582, 15353, 13127, 1543, 8216, 37813, 16262, 115, 8082, 32739, 9567, 226, 1543, 10187, 18390, 2257, 6236, 35125, 6890, 31348, 22381, 40211, 3111, 32739, 9567, 226, 41509, 2289, 1543, 50965], "temperature": 0, "avg_logprob": -0.10707378387451172, "compression_ratio": 1.3148148148148149, "no_speech_prob": 2.066607648898433e-11}, {"id": 15, "seek": 18000, "start": 192.0, "end": 200.0, "text": "下一步你可以點擊AI實錯與糾錯。這是日語的實錯提示詞。你可以添加自己的實錯的規則。", "tokens": [50965, 45271, 31429, 42766, 8216, 37813, 48698, 10376, 13133, 34127, 15239, 122, 13133, 1543, 16600, 6890, 31348, 1546, 10376, 13133, 20949, 25696, 6716, 252, 1543, 42766, 14176, 119, 9990, 17645, 1546, 10376, 13133, 1546, 2888, 237, 46225, 1543, 51365], "temperature": 0, "avg_logprob": -0.10707378387451172, "compression_ratio": 1.3148148148148149, "no_speech_prob": 2.066607648898433e-11}, {"id": 16, "seek": 20000, "start": 200.0, "end": 216.0, "text": "執行完成之後,會出現二次確認的表。檢查後沒有問題。你就可以將常見的錯誤。一鍵批量添加到熱池表中。點擊確認後,程序會自動完成批量的替換。你可以到設置的頁面中,找到熱池表。", "tokens": [50365, 16262, 115, 8082, 41509, 26362, 11, 6236, 7781, 9581, 11217, 9487, 24293, 22041, 1546, 17571, 1543, 40741, 95, 42623, 5661, 6963, 17197, 1543, 2166, 36781, 34168, 11279, 10470, 1546, 13133, 3549, 97, 1543, 2257, 41668, 113, 3416, 117, 26748, 14176, 119, 9990, 4511, 32745, 12800, 254, 17571, 5975, 1543, 8216, 37813, 24293, 22041, 5661, 11, 29649, 6346, 237, 6236, 9722, 12572, 41509, 3416, 117, 26748, 1546, 9531, 123, 36338, 1543, 42766, 4511, 39035, 34719, 1546, 8313, 223, 8833, 5975, 11, 25085, 4511, 32745, 12800, 254, 17571, 1543, 51165], "temperature": 0, "avg_logprob": -0.08523714628150995, "compression_ratio": 1.4505494505494505, "no_speech_prob": 1.988247934348042e-11}, {"id": 17, "seek": 20000, "start": 216.0, "end": 226.0, "text": "切換到日語。你就會看到熱池了。在這裡你可以管理你的熱池。然後在轉錄的時候,開啟AI熱池替換開關就可以了。", "tokens": [51165, 23632, 36338, 4511, 6890, 31348, 1543, 41045, 6236, 18032, 32745, 12800, 254, 2289, 1543, 3581, 18771, 42766, 23131, 13876, 18961, 32745, 12800, 254, 1543, 10213, 3581, 32739, 9567, 226, 20643, 11, 8949, 3284, 253, 48698, 32745, 12800, 254, 9531, 123, 36338, 8949, 14899, 36781, 2289, 1543, 51665], "temperature": 0, "avg_logprob": -0.08523714628150995, "compression_ratio": 1.4505494505494505, "no_speech_prob": 1.988247934348042e-11}, {"id": 18, "seek": 22600, "start": 226.0, "end": 239.0, "text": "最後,有雙語字幕的需求的。可以點擊下雙語字幕。上行可以選擇日語,下行選擇中文。點擊開始處理。等待一會兒,雙語的字幕就生成好了。", "tokens": [50365, 20578, 11, 2412, 6306, 247, 31348, 22381, 40211, 1546, 32535, 32718, 1546, 1543, 6723, 8216, 37813, 4438, 6306, 247, 31348, 22381, 40211, 1543, 5708, 8082, 6723, 21625, 13167, 229, 6890, 31348, 11, 4438, 8082, 21625, 13167, 229, 5975, 17174, 1543, 8216, 37813, 21017, 39289, 13876, 1543, 10187, 18390, 2257, 6236, 35125, 11, 6306, 247, 31348, 1546, 22381, 40211, 3111, 8244, 11336, 12621, 1543, 51015], "temperature": 0, "avg_logprob": -0.11180734634399414, "compression_ratio": 1.3668341708542713, "no_speech_prob": 2.796528633663975e-11}, {"id": 19, "seek": 22600, "start": 241.0, "end": 247.0, "text": "點擊導出,就可以將ASR的字幕導出了。我是浩叔,我們下一個視頻見。", "tokens": [51115, 8216, 37813, 25647, 7781, 11, 36781, 34168, 3160, 49, 1546, 22381, 40211, 25647, 7781, 2289, 1543, 15914, 11038, 102, 39936, 11, 5884, 4438, 8990, 27333, 29940, 10470, 1543, 51415], "temperature": 0, "avg_logprob": -0.11180734634399414, "compression_ratio": 1.3668341708542713, "no_speech_prob": 2.796528633663975e-11}], "language": "zh"}