Suno Introduces Speech Beta: Combining Voice and Music into a Single Model – Unite.AI

Suno Launches Revolutionary Speech Beta: A New Era in Audio Creation

On October 1, 2026, Suno unveiled its latest innovation, Speech (beta). This groundbreaking spoken-audio model is touted as the first to seamlessly blend voice and music into a single, cohesive track. After a month of testing with a select group, the beta version is now accessible to all users on mobile and web platforms.

The announcement was made by Chief Product Officer Jack Brody in a detailed blog post. Speech allows users to create spoken audio paired with original background music. Users simply input their ideas, poems, or texts, and specify the desired voice and musical style, making content creation more intuitive than ever.

Experience the Power of Speech Beta

A release note

from the launch day describes this model as unique in its ability to produce speech and its accompanying soundtrack simultaneously. Available on Android, iOS, and web platforms, the beta showcases potential applications, such as calming bedtime stories paired with soft piano music, motivational speeches backed by energetic stadium drums, and even ASMR grocery lists.

Defining the Future of Creative Entertainment

Suno envisions the launch of Speech as a pivotal step in what they call “creative entertainment,” which they believe will shape the next wave of consumer technology. Music remains at the core of Suno’s offerings, while their vision expands to encompass diverse forms of human expression. The blog post highlights Speech as yet another canvas for the individuality of its community, suitable for various special occasions, emotions, and relationships.

Exploring Early Uses and Noted Limitations

During initial testing, Suno’s team explored the model’s capabilities by transforming friends’ text messages into dramatic recitations and adding cinematic scores to simple voice notes. They also crafted meditative experiences, heartfelt poems, and enchanting bedtime stories for children.

However, Suno cautions that users should expect beta-like performance. For instance, accents may occasionally shift unexpectedly, and some “dramatic pauses” might be quite exaggerated. The company encourages users to find creative uses that may not have been anticipated, emphasizing that this exploration is crucial to opening up the beta.

Recap of Suno’s 2026 Innovations

The rollout of Speech beta follows an exciting array of releases from Suno throughout 2026. On September 9, the company introduced v6, a new generation of music models developed in collaboration with industry giants like Warner Music Group, BMG, and Believe. The v6 models are hailed as Suno’s most advanced yet, delivering faster, more expressive, and higher-quality audio experiences.

What’s New with v6?

Marking a significant upgrade, v6 allows creators to edit portions of existing songs using natural language, create mashups from multiple sources in one go, and even adapt a single lyric without having to reconstruct the entire piece. One example provided involved changing a chorus to feature a gospel choir.

Collaborative Learning and Future Enhancements

Suno engages with artists, producers, and songwriters through weekly writing camps to better understand how they integrate Suno into their creative workflows. The company is attentive to improving user experiences, with ongoing efforts to enhance the platform’s capabilities and introduce safeguards against unauthorized content use.

As they phase out older models, Suno aims to shift entirely to the v6 platform, while also developing personalized opt-in experiences for individual artists, ensuring they are compensated for their contributions.

Leading the Charge in Audio Innovation

In addition to the new Speech feature, Suno previously launched v5.5 on March 26, 2026, introducing features like Voices, Custom Models, and My Taste, empowering users to personalize their audio creations further. The recent updates also included Studio 2.0, a revamped generative audio workstation equipped with advanced features such as MIDI editing and built-in synths.

Suno remains committed to evolving the Speech feature based on community feedback and user experiences, inviting input to guide future improvements.

Here are five frequently asked questions (FAQs) regarding the Suno Launches Speech Beta, pairing voice and music in one model:

FAQ 1: What is the Suno Speech Beta?

Answer: The Suno Speech Beta is an innovative model launched by Suno that combines advanced speech synthesis with music integration. It aims to create a seamless experience where voice and musical elements can be combined effectively, enhancing applications like virtual assistants, audiobooks, and interactive media.

FAQ 2: How does the pairing of voice and music work in this model?

Answer: The model utilizes advanced algorithms to synchronize speech with musical backgrounds, allowing for a harmonious blend. It analyzes the emotional tone and pacing of the spoken content and adjusts the music accordingly to create an engaging audio experience.

FAQ 3: What are the potential applications for this technology?

Answer: This technology can be used in various applications, including interactive storytelling, gaming, podcasting, virtual assistants, and educational tools, where a dynamic audio backdrop can enhance user engagement and retention.

FAQ 4: Is the Suno Speech Beta available for public use?

Answer: As of the launch announcement, the Suno Speech Beta may be available for developers and selected users for testing and feedback. Future updates will likely provide more information on wider accessibility and usage options.

FAQ 5: How does this model differ from traditional text-to-speech systems?

Answer: Unlike traditional text-to-speech systems that focus solely on converting text into spoken words, the Suno Speech Beta integrates musical elements, providing a richer auditory experience. This combination allows for a more nuanced and expressive way of delivering content, adjusting tone and emotion in real time.

Feel free to ask if you need more information!

Source link

CNTXT AI Unveils Munsit: The Most Precise Arabic Speech Recognition System to Date

Revolutionizing Arabic Speech Recognition: CNTXT AI Launches Munsit

In a groundbreaking development for Arabic-language artificial intelligence, CNTXT AI has introduced Munsit, an innovative Arabic speech recognition model. This model is not only the most accurate of its kind but also surpasses major players like OpenAI, Meta, Microsoft, and ElevenLabs in standard benchmarks. Developed in the UAE and designed specifically for Arabic, Munsit is a significant advancement in what CNTXT dubs “sovereign AI”—technological innovation built locally with global standards.

Pioneering Research in Arabic Speech Technology

The scientific principles behind this achievement are detailed in the team’s newly published paper, Advancing Arabic Speech Recognition Through Large-Scale Weakly Supervised Learning. This research introduces a scalable and efficient training method addressing the chronic shortage of labeled Arabic speech data. Utilizing weakly supervised learning, the team has created a system that raises the bar for transcription quality in both Modern Standard Arabic (MSA) and over 25 regional dialects.

Tackling the Data Scarcity Challenge

Arabic, one of the most widely spoken languages worldwide and an official UN language, has long been deemed a low-resource language in speech recognition. This is due to its morphological complexity and the limited availability of extensive, labeled speech datasets. Unlike English, which benefits from abundant transcribed audio data, Arabic’s dialectal diversity and fragmented digital footprint have made it challenging to develop robust automatic speech recognition (ASR) systems.

Instead of waiting for the slow manual transcription process to catch up, CNTXT AI opted for a more scalable solution: weak supervision. By utilizing a massive corpus of over 30,000 hours of unlabeled Arabic audio from various sources, they constructed a high-quality training dataset of 15,000 hours—one of the largest and most representative Arabic speech collections ever compiled.

Innovative Transcription Methodology

This approach did not require human annotation. CNTXT developed a multi-stage system to generate, evaluate, and filter transcriptions from several ASR models. Transcriptions were compared using Levenshtein distance to identify the most consistent results, which were later assessed for grammatical accuracy. Segments that did not meet predefined quality standards were discarded, ensuring that the training data remained reliable even in the absence of human validation. The team continually refined this process, enhancing label accuracy through iterative retraining and feedback loops.

Advanced Technology Behind Munsit: The Conformer Architecture

The core of Munsit is the Conformer model, a sophisticated hybrid neural network architecture that melds the benefits of convolutional layers with the global modeling capabilities of transformers. This combination allows the Conformer to adeptly capture spoken language nuances, balancing both long-range dependencies and fine phonetic details.

CNTXT AI implemented an advanced variant of the Conformer, training it from scratch with 80-channel mel-spectrograms as input. The model consists of 18 layers and approximately 121 million parameters, with training conducted on a high-performance cluster utilizing eight NVIDIA A100 GPUs. This enabled efficient processing of large batch sizes and intricate feature spaces. To manage the intricacies of Arabic’s morphology, they employed a custom SentencePiece tokenizer yielding a vocabulary of 1,024 subword units.

Unlike conventional ASR training that pairs each audio clip with meticulously transcribed labels, CNTXT’s strategy relied on weak labels. Though these labels were less precise than human-verified ones, they were optimized through a feedback loop that emphasized consensus, grammatical correctness, and lexical relevance. The model training utilized the Connectionist Temporal Classification (CTC) loss function, ideally suited for the variable timing of spoken language.

Benchmark Dominance of Munsit

The outcomes are impressive. Munsit was tested against leading ASR models on six notable Arabic datasets: SADA, Common Voice 18.0, MASC (clean and noisy), MGB-2, and Casablanca, which encompass a wide array of dialects from across the Arab world.

Across all benchmarks, Munsit-1 achieved an average Word Error Rate (WER) of 26.68 and a Character Error Rate (CER) of 10.05. In contrast, the best-performing version of OpenAI’s Whisper recorded an average WER of 36.86 and CER of 17.21. Even Meta’s SeamlessM4T fell short. Munsit outperformed all other systems in both clean and noisy environments, demonstrating exceptional resilience in challenging conditions—critical in areas like call centers and public services.

The performance gap was equally significant compared to proprietary systems, with Munsit eclipsing Microsoft Azure’s Arabic ASR models, ElevenLabs Scribe, and OpenAI’s GPT-4o transcription feature. These remarkable improvements translate to a 23.19% enhancement in WER and a 24.78% improvement in CER compared to the strongest open baseline, solidifying Munsit as the premier solution in Arabic speech recognition.

Setting the Stage for Arabic Voice AI

While Munsit-1 is already transforming transcription, subtitling, and customer support in Arabic markets, CNTXT AI views this launch as just the beginning. The company envisions a comprehensive suite of Arabic language voice technologies, including text-to-speech, voice assistants, and real-time translation—all anchored in region-specific infrastructure and AI.

“Munsit is more than just a breakthrough in speech recognition,” said Mohammad Abu Sheikh, CEO of CNTXT AI. “It’s a statement that Arabic belongs at the forefront of global AI. We’ve demonstrated that world-class AI doesn’t have to be imported—it can flourish here, in Arabic, for Arabic.”

With the emergence of region-specific models like Munsit, the AI industry enters a new era—one that prioritizes linguistic and cultural relevance alongside technical excellence. With Munsit, CNTXT AI exemplifies the harmony of both.

Here are five frequently asked questions (FAQs) regarding CNTXT AI’s launch of Munsit, the most accurate Arabic speech recognition system:

FAQ 1: What is Munsit?

Answer: Munsit is a cutting-edge Arabic speech recognition system developed by CNTXT AI. It utilizes advanced machine learning algorithms to understand and transcribe spoken Arabic with high accuracy, making it a valuable tool for various applications, including customer service, transcription services, and accessibility solutions.

FAQ 2: How does Munsit improve Arabic speech recognition compared to existing systems?

Answer: Munsit leverages state-of-the-art deep learning techniques and a large, diverse dataset of Arabic spoken language. This enables it to better understand dialects, accents, and contextual nuances, resulting in a higher accuracy rate than previous Arabic speech recognition systems.

FAQ 3: What are the potential applications of Munsit?

Answer: Munsit can be applied in numerous fields, including education, telecommunications, healthcare, and media. It can enhance customer support through voice-operated services, facilitate transcription for media and academic purposes, and support language learning by providing instant feedback.

FAQ 4: Is Munsit compatible with different Arabic dialects?

Answer: Yes, one of Munsit’s distinguishing features is its ability to recognize and process various Arabic dialects, ensuring accurate transcription regardless of regional variations in speech. This makes it robust for users across the Arab world.

FAQ 5: How can businesses integrate Munsit into their systems?

Answer: Businesses can integrate Munsit through CNTXT AI’s API, which provides easy access to the speech recognition capabilities. This allows companies to embed Munsit into their applications, websites, or customer service platforms seamlessly to enhance user experience and efficiency.

Source link

Advancing Speech Data Collection in Europe for All Languages

The Importance of Language Diversity in AI Development

The world of AI language models has primarily focused on English, leaving many European languages underrepresented. This imbalance has significant implications for how AI technologies interact with various languages and cultures. MOSEL seeks to change this narrative by providing a rich collection of speech data for all 24 official languages of the European Union, promoting inclusivity and diversity in AI development.

Empowering Multilingual AI Models with MOSEL

Language diversity plays a crucial role in ensuring that AI technologies are inclusive and accessible to all. By incorporating multilingual datasets, AI systems can better serve users regardless of the language they speak. Embracing linguistic diversity allows for technology that is more accessible and reflective of the varied needs and cultures of its users.

Discovering MOSEL: A Game-Changer in Multilingual AI Development

MOSEL, Massive Open-source Speech data for European Languages, is a groundbreaking project that aims to provide a comprehensive collection of speech data for all 24 European Union languages. This open-source initiative integrates data from various projects to advance multilingual AI development.

Enhancing Language Models with Transcribed and Unlabeled Data

One of MOSEL’s key contributions is the inclusion of both transcribed and unlabeled data, offering a unique opportunity to develop more inclusive language models. The combination of these datasets allows for a deeper understanding of Europe’s linguistic diversity.

Addressing Data Disparities for Underrepresented Languages

MOSEL strives to bridge the gap in speech data availability for underrepresented languages by leveraging innovative techniques like OpenAI’s Whisper model. By transcribing previously unlabeled audio data, MOSEL expands training material, especially for languages with limited resources.

Championing Open Access for AI Innovation

MOSEL’s open-source approach empowers researchers and developers to work with extensive speech data, fostering collaboration and experimentation in European AI research. This accessibility levels the playing field, allowing smaller organizations and institutions to contribute to cutting-edge AI advancements.

Future Outlook: Advancing Inclusive AI Development with MOSEL

As MOSEL continues to expand its dataset, particularly for underrepresented languages, the project aims to create a more balanced and inclusive resource for AI development. By setting a precedent for inclusivity, MOSEL paves the way for a more equitable technological future globally.

  1. What is the goal of the MOSAIC project?
    The MOSAIC project aims to advance speech data collection for all European languages, ensuring a more diverse and representative dataset for research and development in the field of speech technology.

  2. How does MOSAIC plan to collect speech data for all European languages?
    MOSAIC will leverage crowd-sourcing platforms to engage speakers of various European languages in recording speech data. This approach allows for a large-scale and cost-effective collection process.

  3. Why is it important to have speech data for all European languages?
    Having speech data for all European languages is crucial for developing inclusive and accurate speech technology systems that can cater to a diverse range of users. This ensures that no language is left behind in the advancement of technology.

  4. How can individuals contribute to the MOSAIC project?
    Individuals can contribute to the MOSAIC project by participating in speech data collection tasks on the designated crowd-sourcing platforms. By recording their voices, they can help create a more comprehensive dataset for their respective languages.

  5. What are some potential applications of the speech data collected through MOSAIC?
    The speech data collected through MOSAIC can be used for various applications, including speech recognition, natural language processing, and virtual assistants. By expanding the availability of speech data for all European languages, MOSAIC opens up new possibilities for technological advancements in these areas.

Source link