AI-Powered Language Speaking Practice: An Extensive Review of Top Tools for Learners
Part 1 of a two-part series.
My Colombian mother-in-law has been trying to learn English for about 30 years. She pulls out a large notebook daily and scribbles down phrases and verbs she learns from English-language podcasts, YouTube lessons, or Duolingo sessions.
But she can’t hold a conversation in English. Not even a basic one. Whenever I try to engage her in conversation, she smiles sheepishly and says she’s too shy to speak English. She’s learned many words and phrases, but doesn’t want to make a mistake when speaking. So, she continues each day with her scribblings, content to write down English phrases, but unable to converse with an English speaker.
I don’t mean to pick on my sweet mother-in-law. She’s certainly not alone in the fear of making mistakes. My stepdaughter taught English in Japan for a year, and her students were extremely reluctant to speak English, despite their high English vocabulary and grammar proficiency. She came away convinced that Japanese students were simply too afraid of making mistakes.
I understand all this personally. I grew up in an English-speaking suburb of Montreal. I took French classes as a kid and heard plenty of French around me, but I was shy and nervous about speaking in public. As a teenager, it was more about resistance. The Quebec government at the time was restricting or banning the use of English in various situations, and I resented their policies.
Years later, however, I let go of any resentment and fear of making mistakes and spoke French regularly at work and with friends. It served me well later during my four years living and working in France and Switzerland. I relocated to the U.S., and some 15 years afterward, a friend asked me to join him in a beginner’s Spanish class. I soon became enamored with learning Spanish, decided to travel around South America, and eventually married a wonderful Colombiana. (We are both studying Italian.)
Learning to speak other languages has transformed my life. It’s opened me up to people, cultures, and opportunities in ways I never dreamed of and has given me a deeper understanding of the world. But it only happened once I learned not to fear making mistakes.
Language teachers will tell you that if you want to speak a new language … speak the language. But often, students are uncomfortable speaking in front of their peers and, in any event, there is relatively little class time available each day to practice speaking. Furthermore, language partners may not be readily available to students.
Fortunately, in recent years, a new generation of AI-powered language learning tools has emerged to help students practice speaking in world languages. These tools leverage advanced AI conversation models (like GPT-4 and Google’s Gemini) to simulate realistic dialogues, give feedback, and personalize practice. Unlike traditional apps focusing on drills or vocabulary, AI tutors can engage learners in open-ended conversations – from ordering coffee in Spanish to debating a topic in French – all in a low-pressure, private environment. Many of these tools are readily accessible via mobile apps or web platforms and many are affordable — freemium or under ~$10/month.
In this two-part series, I’ll analyze prominent AI-infused world language speaking tools – Duolingo Max, Talkpal, Langua, ChatGPT (with voice) – and identify other notable newcomers. I’ll examine how each tool supports spoken practice and evaluate their pedagogical benefits for students. I’ll compare their strengths and weaknesses, based on my personal experience and commentaries from an assortment of users.
Educational Benefits of AI-powered Speaking Tools
For starters, AI-powered language tools offer a number of educational benefits that align with modern language-learning pedagogy. Perhaps the biggest benefit is providing massive amounts of speaking practice in a way that is personalized, private, and pressure-free. In a typical school environment, a student might only speak a few minutes of, say, French per class, and might feel self-conscious about errors. With an AI partner, that same student can practice speaking French for 20 minutes at home, freely making mistakes – the AI will patiently continue the conversation and gently correct them, without any embarrassment. This kind of low-affective-filter environment is something language teachers have long sought, because when anxiety is low, students are more likely to experiment with the language and thus learn more effectively.
Personalization and adaptive learning: AI tutors can adapt to each learner’s level and pace. Unlike a classroom dialog where one size often fits all, an AI can simplify its speech or repeat things as needed for a struggling learner, or ramp up the complexity for an advanced learner. For example, Talkpal and Langua both adjust the difficulty of vocabulary and sentences based on the user’s proficiency and even within a session, if the AI senses the user is comfortable, it might introduce more advanced phrases. This ties into personalized instruction, rather than a fixed curriculum, the AI essentially creates a customized lesson on the fly, reacting to the learner’s performance. Additionally, many AIs allow the user to choose topics of interest (sports, music, travel plans) so the conversation is relevant and engaging for them.
Multimodal learning: These tools often engage multiple skills at once. For instance, when using an AI like Duolingo Max, the student is speaking and listening (auditory), reading the AI’s transcribed replies (visual text), sometimes clicking to see translations or images, and then possibly reviewing flashcards (memory reinforcement). This multimodal approach can lead to better retention. An example is the integration of images or avatars – Duolingo’s roleplays have cute graphics and characters that create a story, making the learning experience more memorable. (Duolingo Stories are often hilarious.) Google’s Little Language Lessons experiments also hint at using the camera (Word Cam) to learn words by sight. So we see a trend of combining modalities: not just a blank chat screen, but maybe an image of a café, plus a conversation, plus text feedback. This caters to different learning preferences and keeps students engaged.
Feedback and corrective support: AI tutors provide instant feedback in a way that textbooks or even recorded videos cannot. For example, many of these tools will immediately highlight a grammar error and show the correct form, or will prompt you to retry a sentence you mispronounced. This immediate loop of action→feedback→adjustment is ideal for skill acquisition. Tools like Langua give comprehensive post-conversation feedback (written and audio), which a motivated student can use to self-correct and study further. It’s worth noting, though, that feedback quality varies, and teachers should guide students on interpreting it.
Confidence and fluency building: Because AI partners remove the fear of judgment, students can practice speaking more freely, which builds fluency over time. As several users attest, doing daily chats with AI improves speaking fluidity and confidence. If AI lets a student actually speak more, that alone can yield improvements. For shy students, practicing privately first can prepare them to speak up more in class. The AI serves as a stepping stone to real interactions. By the time a student goes to converse with a native speaker, they’ve already rehearsed common conversation patterns with AI multiple times, making them more at ease.
Realistic avatars and role-play: The use of avatars (like Duolingo’s Lily or the characters in Talkpal) and role-play scenarios adds an element of engagements and context that abstract exercises lack. Role-play is a proven technique in language pedagogy, and AI enables infinite role-play partners without needing classmates or teachers to play every role. Some tools include visual avatars – Duolingo shows Lily’s face, Univerbal uses AI images of people, etc. Future tools might even animate those avatars realistically. This can help with paralinguistic awareness: if an avatar shows a confused face when you say something incorrectly, that could cue the learner that they weren’t understood (this isn’t implemented widely yet, but it’s conceivable). Even without actual animated cues, the very idea that you are “talking to Lily” or “chatting with a friend at a café” can shift the mindset from exercise to real communication, which is pedagogically beneficial. It taps into narrative and situational learning – people often remember language better when it’s learned in the context of a story or scenario, as opposed to isolated drills.
Practice for specific purposes: AI tools can be tailored to practice for specific scenarios like job interviews, presentations, or travel. This is great for older students or those with clear goals. For example, an AP French student can practice the AP oral exam prompts with ChatGPT acting as the exam recorder. In the past, they might have had to hire a tutor for this or impose on a fluent friend; now AI can fill that role affordably.
Duolingo Max: AI Voice Chats and Role Playing in a Gamified Environment
Starting our tour with Duolingo, the world’s most popular learning app, makes sense. I’ve used it for Spanish and Italian for years, my wife uses it to improve her English, and I have nephews using it to learn English and Portuguese. In 2023, Duolingo introduced “Max,” a premium tier powered by GPT-4, adding three main features: Explain My Answer, Roleplay, and Video Call with Lily. This “Call Lily” feature is the only one available for a sustained verbal conversation and is the focus of this review. While Duolingo Max is costly, it’s worth examining because it brings AI-driven conversation practice into a mainstream app in a very student-friendly format.
In partnership with Essay Grader
Essay Grader is a straightforward, powerful tool that uses AI to take the load off your shoulders and will completely change the way you approach grading!
Every feature puts teachers first, with a focus on simplifying repetitive tasks like grading to improve productivity - all the while letting teachers be fully in control.
Streamline grading, provide detailed, actionable feedback to your students, and maintain complete control over your assessments - all in one intuitive interface.
Reduce grading time by 80%. Grade your entire class's essays in 2 minutes or less and deliver high-quality, specific feedback to your students.
Ensure fair essay grading by removing bias and aligning with state standards like Texas STAAR, Florida BEST and California CCSS.
Join Essay Grader and its ever-expanding network of 60,000+ educators, and you’ll connect with others who share your passion and commitment.
AI voice calls (“Call Lily”): This feature is like a mini video call, except the conversation is with a cartoon character called Lily, a sarcastic (but amusing) teen powered by AI. In the call, Lily appears animatedly and talks to you in the target language. You hear her through your device, and you respond by speaking.
The call lasts about two minutes before Lily wraps up (she often says she must go). The calls are frequent, typically 1-2 calls within a fifteen-minute set of Duolingo exercises. Lily usually dictates the direction of the conversation, asking the user if it’s “okay” to talk about the subject she has in mind. And you need be brief in your responses to her questions. Lily will interrupt you when you speak too long.
Despite the brevity and interruptions, many users who have tried it are impressed: “The voice recognition is much better than [in regular] Duolingo’s exercises,” one user said of the Japanese version. Another user reported: “In just a few weeks, I feel my speaking skills have advanced massively… the video call feature is a perfect fix [for speaking practice].” My wife likewise feels empowered through her conversations with Lily, though she doesn’t appreciate the interruptions.
Don’t put me in this group of happy users. Frankly, I’m disappointed with “Call Lily” and often frustrated with the AI. For one, the calls are too brief. Two minutes is too short for an intermediate or advanced speaker, and Lily’s repetitive greeting and goodbye sequences eat into that time. The conversational content is also repetitive. Most calls gravitate to talking about yourself (pet, entertainment, family), which gets bland quickly. These feel like beginner topics and I often find I’m not challenged. It also annoys me when Lily cuts me off. Each time she asks me a question, I immediately think about how to give a short 1-2 sentence response to avoid being cut off.
I also have persistent technical problems. I have an older iPhone, iOS 16.7.11, and invariably there’s a long delay in the middle of each conversation. Lily can’t “hear” a response and keeps waiting for it. It’s no use for me to keep repeating what I’ve just said. I must wait roughly 45 seconds before Lily can process my response. That’s frustrating, especially since the call is so short.
(Lily’s personality on these calls is much more boring than her scripted persona. Instead of witty sarcasm, “Call Lily” is polite and patient, which feels off-character —even if it’s ideal for an AI tutor.)
Duolingo provides a conversation transcript immediately after the call with suggestions on improving your responses. However, the transcript often contains errors because it incorrectly identifies words or phrases I used in the conversation. This negatively impacts the quality of the recommendations and the transcripts become of little use to me.
Value and limitations: Duolingo Max costs roughly $30/month, which is a steep jump from the ~$7/month Super plan. For that price, the added value comes down to the AI features. Are they worth it? Many reviewers say not yet. The French Together blog notes that Duolingo Max is fun, but it’s not a full conversation experience. I agree.
Duolingo Max’s usefulness lies in bringing AI into a structured curriculum. For schools or parents already using Duolingo as a supplement, Max provides a way to practice conversation and get instant help within one app. It’s also a very safe environment: talking to a cartoon character (with strong AI safety guardrails) feels less intimidating than talking to a faceless AI tutor. The familiar faces and gamified rewards can also motivate teens to try the conversations where they might otherwise avoid spontaneous speaking.
“Call Lily” clearly adds a new dimension of speaking and conversational practice to the world’s most-used language app, with clear pedagogical intent (contextual practice, immediate feedback). Overall, the Max features are well-aligned to themes of multimodal learning (combining text, speech, and even visual avatars) and personalized instruction (AI explanations tailored to the user’s answer).
Yet, given the cost, I’d opt for alternatives to practice speaking. I’d only recommend Duolingo Max if a student is already heavily invested in Duolingo. “Call Lily” does provide excellent quick-speaking practice perfect for an intermediate student to do each night, but it offers little else in terms of a full and free-flowing conversation.
My Grade: B-
Talkpal: Free-Flowing, AI-Powered Speaking Practice
Talkpal is a popular AI language tutor app that immerses learners in conversation and speaking exercises across dozens of languages. Talkpal’s interface is inviting and beginner-friendly, featuring a range of learning modes: a free-form Chat mode with an AI tutor, structured Roleplay scenarios (e.g. ordering food at a restaurant), Debates on various topics, a Characters mode (talk to historical or fictional personas), and even a Photo/Caption mode where you describe a picture.
Talkpal’s strength is letting you practice speaking freely in conversations — with “Call Mode”. It’s AI conversationalist “Emma” is great — patient, postive and upbeat. She also asks probing questions that can take conversations in unanticipated, but engaging, directions. And Emma won’t let you stop talking. (You: “This has been a helpful conversation, but it’s late and I’m going to bed now.” Talkpal: “I understand and hope you have a nice rest. So, tell me, what are your plans for tomorrow?”) There are also immediate transcripts with translations. Finally, you can adjust “voice speed” in Settings, so if you want Emma to speak a little slower or faster, you can do so. (You can’t do this in Duolingo.)
Talkpal’s speech recognition and language model appear explicitly tuned for language learners. I’m struck, for instance, by how much better Talkpal’s Emma understands me than Duolingo’s Lily. With Talkpal, I feel like I have a conversation partner on demand. As a teacher, I can assign students to role-play a scenario with Talkpal (e.g. “talk to the AI at the train station to buy a ticket”) as homework. Students would get instant practice and some feedback.
After a conversation with Emma, you’ll receive a “practice summary” with a score and a detailed analysis of your speaking session. You’ll get pointed feedback, usually a paragraph each on issues such as: verb conjugation and tense consistency, pronoun usage and object placement, use of prepositions, vocabulary precision and articles, and sentence structure and connectors.
Many learners who are too shy to speak with classmates or tutors have found Talkpal to be a “perfect” solution to start speaking without fear. Users (like me) generally love the experience of practicing with Talkpal. (The app has a 4.7★ rating on the App Store with over a million downloads.) But there are reports of Talkpal failing to catch certain grammar mistakes and even occasionally correcting things that weren’t wrong.
Talkpal also fails to move you along a structured course curriculum the way that Duolingo does. It offers plenty of valuable and varied exercises, but no clear direction on how to structure your learning. In contrast, Duolingo wants you to follow a particular learning trajectory. In that vein, “Call Lily” often focuses conversations on topics recently covered by Duolingo. Talkpal’s Emma will ask you questions to start a conversation, but essentially lets you dictate the direction of the conversation.
Call Mode and other “premium” Talkpal features are available for $72 a year, an affordable $6 per month, and is considerably cheaper than Duolingo Max.
Overall, Talkpal’s utility lies in providing endless conversational practice in a fun, low-stakes way. It excels at engagement and breadth of content (you won’t run out of things to talk about), though it falls short as an “accurate coach” in some grammatical aspects aspects. Students can improve fluency and comfort with speaking by using Talkpal, as long as they remain aware that not every correction is perfect.
My Grade: A-
Langua: Immersive AI Tutor with Realistic Conversations
Langua has quickly gained a reputation as one of the most advanced AI language tutors for speaking fluency. Launched in 2024, it combines state-of-the-art conversation AI with features designed specifically for language learners. The core of Langua is its AI conversational partner, which users describe as uncannily human-like. When you speak to Langua’s tutor, it responds with an actual voice that carries emotion and native-like pronunciation – learners say they “feel like [they’re] having a real conversation with a native speaker”. This immersion is further enhanced by choosing tutor personas: you can select from various AI characters/voices (e.g. “Carlos, a tutor with a Mexican accent”) to practice regional dialects or just pick a voice that you like. Such personalization can help students attune their ear to the specific accent (useful for AP French vs. Canadian French, for instance).
Speaking experience: Conversations with Langua are very flexible. You can dive into open-ended chat (“Chat about anything”) or choose specific conversation options like Role-plays, Debates, Games, or even targeted grammar practice. The AI adapts to your level; at the start, you set your approximate CEFR level (A1 to C1+), and it will adjust its vocabulary and sentence complexity accordingly. During a conversation, if you get stuck, you can ask the AI for help (in English or the target language), and it will gently assist, much like a kind teacher. One reviewer testing Langua noted he could interrupt the AI to ask grammar questions mid-conversation, and it would break character to explain politely before resuming the dialogue.
Yet, a user must deliberately activate “call mode” for verbal-driven conversation. By default, most interactions on Langua are text-based, so the user must go into Chat Settings and specifically select “Call mode (Beta)” to initiate a verbal exchange. Once the conversation begins, Langua displays a real-time transcription of the AI’s spoken responses. While helpful for understanding, this transcription can be counterproductive for developing real-life auditory and verbal skills. To counter this, you can tap the “view/hide” (eye) icon to remove the text from view and focus purely on listening.
Feedback and learning features: Where Langua truly shines is its suite of feedback tools surrounding the conversation. It offers real-time corrections. For example, if you make a mistake, the AI might subtly correct you in its next reply or flash a written correction on screen without interrupting the flow. After the conversation, Langua provides a written summary highlighting new vocabulary you encountered (with definitions) and any grammar points or mistakes for review.
Langua also addresses pronunciation and speaking fluency. While it may not grade your accent like Talkpal, it does sometimes provide alternative phrasing to sound more native-like and can give tips if you consistently mispronounce something. The AI voices themselves set a high bar for pronunciation, by shadowing or imitating the AI, students can improve their accent. Some users have reported noticeable improvements in fluency and a reduction in choppy speech by regularly talking with Langua.
User and expert feedback: I don’t have nearly the experience with Langua that I have with Duolingo and Talkpals. But I can say that Lengua has an impressive array of integrated features for verbal and non-verbal conversations, including the ability to choose speaking partners based on dialects and speed. The AI speaking partners do indeed sound eerily human and they are adept at providing concrete feedback.
Overall, Lengua has received glowing feedback from both learners and independent reviewers. It boasts a 4.6–4.8★ average rating on review sites. Users seem to agree that Langua is fantastic for unlimited practice and confidence-building, but that humans still provide intangible benefits (Note that “LanguaTalk” from Lengua is a live human tutoring service featuring paid teacher sessions).
Langua’s development team seems quite active in improving their product. As of mid-2025 they have added features (phone-calls mode, beginner lessons, more voices) and appear responsive to user feedback. They admit the AI is “imperfect” and occasionally makes mistakes, but claim to be continually refining it. Data privacy is also emphasized: conversations are stored locally and not reused elsewhere, so students (or parents) need not worry about info leaking.
Bottom line: My takeaway is that Langua is an excellent tool for intermediate to advanced students aiming to achieve fluency or improve their speaking naturalness. (I think beginners are better served by Duolingo.) Lengua fills a gap between casual app practice and an immersive conversation with a native speaker. High school or college students might use Langua to practice for oral exams or just to get more speaking reps beyond the few minutes they might speak during class.
The main downsides are cost — it’s $25 a month — and the fact that verbal-only conversations need be “set”. Yet, for those who can afford it or schools that could subsidize it, Langua offers an exciting glimpse at how AI can provide personalized, engaging, and robust conversational practice.
My Grade: B+
Part II next week…
Any suggestions to improve this newsletter? Please message me or leave a comment below!
Find my book AI Tools & Uses: A Practical Guide for Teachers at Amazon
*Learn more about Essay Grader