Uploaded July 2026 | Updated September 2026, 1 week ago
Universal 3.5 Pro: Smarter Speech-to-Text with Contextual Awareness 🧠🎙️
In this demo, we walk through the new contextual awareness capabilities of Universal 3.5 Pro—and how they dramatically improve transcription accuracy, especially for voice agents.
What's covered:
🔹 Contextual prompting: Pass a description of your audio ("cardiology consultation about chest pain") and accuracy improves within that domain. The more specific the prompt, the better the results.
🔹 Key terms with context: Key terms are no longer applied blindly. Tell the model what a term is ("The user's name is Zachary Klebanoff") and it won't force the term onto acoustically similar phrases.
🔹 Dynamic mid-call prompt updates: Update the prompt in real time via the API (e.g., after a tool call identifies the caller) to boost accuracy as the conversation unfolds.
🔹 Conversation context (agent context): The model retains previous transcriptions in the session and accepts your voice agent's LLM-generated responses as context. If the agent asks "What's your email?", the model expects an email next — without over-biasing. We've seen significant word error rate reductions on voice agent datasets, even in poor audio conditions.
Watch the full demo, including a live customer support scenario with dynamic context updates and mid-call tool calls.
#SpeechToText #VoiceAI #VoiceAgents #ASR #AI
▬▬▬▬▬▬▬▬▬▬▬▬ CONNECT ▬▬▬▬▬▬▬▬▬▬▬▬
🖥️ Website: assemblyai.com
🐦 Twitter: twitter.com/AssemblyAI
🦾 Discord: discord.gg/Cd8MyVJAXd
▶️ Subscribe: youtube.com/c/AssemblyAI?sub_confirmation=1
🔥 We're hiring! Check our open roles: assemblyai.com/careers
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
#MachineLearning #DeepLearning
Universal 3.5 Pro: Smarter Speech-to-Text with Contextual Awareness 🧠🎙️
In this demo, we walk through the new contextual awareness capabilities of Universal 3.5 Pro—and how they dramatically improve transcription accuracy, especially for voice agents.
What's covered:
🔹 Contextual prompting: Pass a description of your audio ("cardiology consultation about chest pain") and accuracy improves within that domain. The more specific the prompt, the better the results.
🔹 Key terms with context: Key terms are no longer applied blindly. Tell the model what a term is ("The user's name is Zachary Klebanoff") and it won't force the term onto acoustically similar phrases.
🔹 Dynamic mid-call prompt updates: Update the prompt in real time via the API (e.g., after a tool call identifies the caller) to boost accuracy as the conversation unfolds.
🔹 Conversation context (agent context): The model retains previous transcriptions in the session and accepts your voice agent's LLM-generated responses as context. If the agent asks "What's your email?", the model expects an email next — without over-biasing. We've seen significant word error rate reductions on voice agent datasets, even in poor audio conditions.
Watch the full demo, including a live customer support scenario with dynamic context updates and mid-call tool calls.
#SpeechToText #VoiceAI #VoiceAgents #ASR #AI
▬▬▬▬▬▬▬▬▬▬▬▬ CONNECT ▬▬▬▬▬▬▬▬▬▬▬▬
🖥️ Website: assemblyai.com
🐦 Twitter: twitter.com/AssemblyAI
🦾 Discord: discord.gg/Cd8MyVJAXd
▶️ Subscribe: youtube.com/c/AssemblyAI?sub_confirmation=1
🔥 We're hiring! Check our open roles: assemblyai.com/careers
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
#MachineLearning #DeepLearning
