{"id":21860,"date":"2025-08-07T11:55:59","date_gmt":"2025-08-07T07:55:59","guid":{"rendered":"https:\/\/krisp.ai\/blog\/?p=21860"},"modified":"2025-08-13T17:11:09","modified_gmt":"2025-08-13T13:11:09","slug":"introducing-krisp-accent-conversion-v3-7","status":"publish","type":"post","link":"https:\/\/krisp.ai\/blog\/introducing-krisp-accent-conversion-v3-7\/","title":{"rendered":"Introducing Krisp Accent Conversion v3.7"},"content":{"rendered":"

Krisp Accent Conversion v3<\/strong>, released in March 2025, marked a breakthrough moment in the evolution of our accent conversion technology. For the first time in two years, we felt the system was mature enough for wide-scale production use.<\/p>\n

 <\/p>\n

In May 2025, we released Accent Conversion v3.5<\/strong>, bringing a major quality upgrade \u2014 with ~20% improvement<\/strong> across key metrics for both Filipino and Indian accents<\/strong> (details here<\/a>). Thanks to Krisp desktop application\u2019s auto-update mechanism, the rollout reached 95% of users within 2 days<\/strong>, and the feedback was overwhelmingly positive, both from agents and customers, driving sentiment and business KPIs.<\/p>\n

 <\/p>\n

In July 2025, we expanded the offering to support the Latin American accent<\/strong> pack. The launch quickly gained traction with several large customers and is now deployed across thousands of agents.<\/p>\n

 <\/p>\n

Throughout this period, we\u2019ve worked closely with partners, agents, and customers to deeply understand corner cases \u2014 especially for the Indian accent, which is the most challenging due to its vast regional variation and phonetic complexity. This close collaboration, combined with relentless efforts from the world-class research and engineering teams at Krisp, has culminated in another major step forward now.<\/p>\n

 <\/p>\n

Today, we\u2019re launching Accent Conversion v3.7<\/strong>, delivering significant improvements in naturalness<\/strong> and voice stability<\/strong>. This release is currently focused on the Indian accent pack, with support for other accents rolling out soon.<\/p>\n

The following sections summarize the key improvements, benchmarking methodology, and a side-by-side comparison of Accent Conversion v3.7 with v3.5.<\/p>\n

<\/p>\n

Key Improvements in AC v3.7<\/h2>\n
    \n
  1. Naturalness: <\/strong>The converted speech sounds even more human-like and natural, with much improved filler-sound handling. Here, expert-rated naturalness scores improved by +14%<\/strong>. Crowdsourced evaluations confirm it with a +6%<\/strong> gain.<\/li>\n
  2. Voice Stability:<\/strong> Enhanced consistency in pitch and tone throughout the utterance, helping avoid unnatural fluctuations, especially for thick accents. This contributed to improved naturalness and clarity scores across all metrics.<\/li>\n
  3. Speech & Audio Clarity:<\/strong> Improvements were noted in both intelligibility and the reduction of artifacts and distortions. Speech Clarity scores rose by 5% in expert assessments, with corresponding enhancements across Meta metrics<\/a>.<\/li>\n
  4. Pronunciation Accuracy:<\/strong> There\u2019s a gain in objective metrics as well, about a 4% relative improvement in Phoneme Error Rate (PER)<\/a>, which can be attributed to more conversational data inclusion in the training. Here, some noticeable accent-specific enhancements in phoneme pronunciation, such as more native-like articulation of \u201cR\u201d and \u201cL\u201d, contribute to a +5% increase in the Accent Conversion score.<\/li>\n<\/ol>\n

    <\/p>\n

    Evaluation Results<\/h2>\n

    For subjective and objective evaluations, 78 real-world recordings were sampled.<\/p>\n

    For the crowdsourced evaluation, each recording received exactly 30 independent votes to ensure statistical confidence, 2340 total votes.<\/p>\n

    The results shown in the table below represent aggregated averages across all recordings.<\/p>\n\n\n\n\n\n\n\n\n\n\n\n
    Metric<\/strong><\/th>\nIN AC v3.5<\/strong><\/th>\nIN AC v3.7<\/strong><\/th>\nComment<\/strong><\/th>\n<\/tr>\n<\/thead>\n
    Expert Evaluation – Natural speech <\/span>(1 to 5)<\/span><\/td>\n3.7<\/td>\n\"\"4.2 (+14%)<\/strong><\/td>\nSpeech sounds even more human-like, with much improved filler-sound handling<\/td>\n<\/tr>\n
    Expert Evaluation – Speech Clarity <\/span>(1 to 5)<\/span><\/td>\n4.0<\/td>\n\"\"4.2 (+5%)<\/strong><\/td>\nSpeech is with fewer artifacts and clearer, especially in slurred and mumbling segments<\/td>\n<\/tr>\n
    Expert Evaluation – Accent Conversion <\/span>(1 to 5)<\/span><\/td>\n4.3<\/td>\n\"\"4.5 (+5%)<\/strong><\/td>\nAccent-specific enhancements in phoneme pronunciation, such as more native-like articulation of \u201cR\u201d and \u201cL\u201d<\/td>\n<\/tr>\n
    Crowdsourced Evaluation<\/strong> – <\/span>\u201cHow natural does the voice sound?\u201d <\/span><\/em>(1 to 5)<\/span><\/td>\n3.4<\/td>\n\"\"3.6 (+6%)<\/strong><\/td>\n78 real-world audio recordings assessed by 30 participants<\/td>\n<\/tr>\n
    Crowdsourced Models\u2019 Comparison<\/strong> – <\/span>Which option sounds more natural?<\/span><\/td>\n1242<\/td>\n1878 <\/span>(+20%)<\/strong><\/td>\n78 real-world audio recording pairs were evaluated, with each pair assessed by 40 participants<\/td>\n<\/tr>\n
    Meta Aesthetic – Natural speech (1 to 10)<\/span><\/td>\n5.6<\/td>\n\"\"5.8 (+4%)<\/strong><\/td>\n<\/td>\n<\/tr>\n
    Meta Aesthetic – Speech Clarity (1 to 10)<\/span><\/td>\n7.5<\/td>\n\"\"7.6 (+1%)<\/strong><\/td>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

     <\/p>\n

    Comparative audio samples<\/strong><\/h2>\n

    Listening Tip:<\/strong> For the most accurate and immersive comparison between v3.5 and v3.7 Accent Conversion, we recommend using quality headphones.<\/p>\n

    This helps highlight the improvements in clarity, naturalness, and speaker identity preservation that may be less perceptible on laptop or mobile speakers.<\/p>\n\n\n\n\n\n\n\n\n\n\n\n\n
    #<\/strong><\/th>\nImprovement Category<\/th>\nOriginal<\/strong><\/th>\nConverted AC v3.5<\/strong><\/th>\nConverted AC v3.7<\/strong><\/th>\n<\/tr>\n<\/thead>\n
    1<\/td>\nSpeech Naturalness<\/td>\n
    2<\/td>\nSpeech Naturalness<\/td>\n
    3<\/td>\nSpeech Naturalness
    \nSpeech Clarity<\/td>\n
    4<\/td>\nSpeech Clarity<\/td>\n
    5<\/td>\nSpeech Clarity
    \nSpeech Naturalness
    \nVoice Stability<\/td>\n
    6<\/td>\nSpeech Clarity
    \nSpeech Naturalness
    \nVoice Stability<\/td>\n
    7<\/td>\nSpeech Naturalness
    \nSpeech Clarity<\/td>\n
    8<\/td>\nSpeech Naturalness
    \nSpeech Clarity<\/td>\n